File size: 127,788 Bytes
9375c9a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 |
<html><!-- Created using the cpp_pretty_printer from the dlib C++ library. See http://dlib.net for updates. --><head><title>dlib C++ Library - loss_abstract.h</title></head><body bgcolor='white'><pre>
<font color='#009900'>// Copyright (C) 2015 Davis E. King ([email protected])
</font><font color='#009900'>// License: Boost Software License See LICENSE.txt for the full license.
</font><font color='#0000FF'>#undef</font> DLIB_DNn_LOSS_ABSTRACT_H_
<font color='#0000FF'>#ifdef</font> DLIB_DNn_LOSS_ABSTRACT_H_
<font color='#0000FF'>#include</font> "<a style='text-decoration:none' href='core_abstract.h.html'>core_abstract.h</a>"
<font color='#0000FF'>#include</font> "<a style='text-decoration:none' href='../image_processing/full_object_detection_abstract.h.html'>../image_processing/full_object_detection_abstract.h</a>"
<font color='#0000FF'>namespace</font> dlib
<b>{</b>
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='EXAMPLE_LOSS_LAYER_'></a>EXAMPLE_LOSS_LAYER_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
A loss layer is the final layer in a deep neural network. It computes the
task loss. That is, it computes a number that tells us how well the
network is performing on some task, such as predicting a binary label.
You can use one of the loss layers that comes with dlib (defined below).
But importantly, you are able to define your own loss layers to suit your
needs. You do this by creating a class that defines an interface matching
the one described by this EXAMPLE_LOSS_LAYER_ class. Note that there is no
dlib::EXAMPLE_LOSS_LAYER_ type. It is shown here purely to document the
interface that a loss layer must implement.
A loss layer can optionally provide a to_label() method that converts the
output of a network into a user defined type. If to_label() is not
provided then the operator() methods of add_loss_layer will not be
available, but otherwise everything will function as normal.
Finally, note that there are two broad flavors of loss layer, supervised
and unsupervised. The EXAMPLE_LOSS_LAYER_ as shown here is a supervised
layer. To make an unsupervised loss you simply leave out the
training_label_type typedef and the truth iterator argument to
compute_loss_value_and_gradient().
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#009900'>// In most cases training_label_type and output_label_type will be the same type.
</font> <font color='#0000FF'>typedef</font> whatever_type_you_use_for_training_labels training_label_type;
<font color='#0000FF'>typedef</font> whatever_type_you_use_for_outout_labels output_label_type;
<b><a name='EXAMPLE_LOSS_LAYER_'></a>EXAMPLE_LOSS_LAYER_</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- EXAMPLE_LOSS_LAYER_ objects are default constructable.
!*/</font>
<b><a name='EXAMPLE_LOSS_LAYER_'></a>EXAMPLE_LOSS_LAYER_</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> EXAMPLE_LOSS_LAYER_<font color='#5555FF'>&</font> item
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- EXAMPLE_LOSS_LAYER_ objects are copy constructable.
!*/</font>
<font color='#009900'>// Implementing to_label() is optional.
</font> <font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
requires
- SUBNET implements the SUBNET interface defined at the top of
layers_abstract.h.
- input_tensor was given as input to the network sub and the outputs are
now visible in layer<i>(sub).get_output(), for all valid i.
- input_tensor.num_samples() > 0
- input_tensor.num_samples()%sub.sample_expansion_factor() == 0.
- iter == an iterator pointing to the beginning of a range of
input_tensor.num_samples()/sub.sample_expansion_factor() elements. Moreover,
they must be output_label_type elements.
ensures
- Converts the output of the provided network to output_label_type objects and
stores the results into the range indicated by iter. In particular, for
all valid i, it will be the case that:
*(iter+i/sub.sample_expansion_factor()) is populated based on the output of
sub and corresponds to the ith sample in input_tensor.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
requires
- SUBNET implements the SUBNET interface defined at the top of
layers_abstract.h.
- input_tensor was given as input to the network sub and the outputs are
now visible in layer<i>(sub).get_output(), for all valid i.
- input_tensor.num_samples() > 0
- input_tensor.num_samples()%sub.sample_expansion_factor() == 0.
- for all valid i:
- layer<i>(sub).get_gradient_input() has the same dimensions as
layer<i>(sub).get_output().
- layer<i>(sub).get_gradient_input() contains all zeros (i.e.
initially, all input gradients are 0).
- truth == an iterator pointing to the beginning of a range of
input_tensor.num_samples()/sub.sample_expansion_factor() elements. Moreover,
they must be training_label_type elements.
- for all valid i:
- *(truth+i/sub.sample_expansion_factor()) is the label of the ith sample in
input_tensor.
ensures
- This function computes a loss function that describes how well the output
of sub matches the expected labels given by truth. Let's write the loss
function as L(input_tensor, truth, sub).
- Then compute_loss_value_and_gradient() computes the gradient of L() with
respect to the outputs in sub. Specifically, compute_loss_value_and_gradient()
assigns the gradients into sub by performing the following tensor
assignments, for all valid i:
- layer<i>(sub).get_gradient_input() = the gradient of
L(input_tensor,truth,sub) with respect to layer<i>(sub).get_output().
Note that, since get_gradient_input() is zero initialized, you don't
have to write gradient information to layers that have a zero
loss gradient.
- returns L(input_tensor,truth,sub)
!*/</font>
<b>}</b>;
std::ostream<font color='#5555FF'>&</font> <b><a name='operator'></a>operator</b><font color='#5555FF'><</font><font color='#5555FF'><</font><font face='Lucida Console'>(</font>std::ostream<font color='#5555FF'>&</font> out, <font color='#0000FF'>const</font> EXAMPLE_LOSS_LAYER_<font color='#5555FF'>&</font> item<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
print a string describing this layer.
!*/</font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_xml'></a>to_xml</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> EXAMPLE_LOSS_LAYER_<font color='#5555FF'>&</font> item, std::ostream<font color='#5555FF'>&</font> out<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
This function is optional, but required if you want to print your networks with
net_to_xml(). Therefore, to_xml() prints a layer as XML.
!*/</font>
<font color='#0000FF'><u>void</u></font> <b><a name='serialize'></a>serialize</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> EXAMPLE_LOSS_LAYER_<font color='#5555FF'>&</font> item, std::ostream<font color='#5555FF'>&</font> out<font face='Lucida Console'>)</font>;
<font color='#0000FF'><u>void</u></font> <b><a name='deserialize'></a>deserialize</b><font face='Lucida Console'>(</font>EXAMPLE_LOSS_LAYER_<font color='#5555FF'>&</font> item, std::istream<font color='#5555FF'>&</font> in<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
provides serialization support
!*/</font>
<font color='#009900'>// For each loss layer you define, always define an add_loss_layer template so that
</font> <font color='#009900'>// layers can be easily composed. Moreover, the convention is that the layer class
</font> <font color='#009900'>// ends with an _ while the add_loss_layer template has the same name but without the
</font> <font color='#009900'>// trailing _.
</font> <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> EXAMPLE_LOSS_LAYER <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>EXAMPLE_LOSS_LAYER_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font><font color='#009900'>// ----------------------------------------------------------------------------------------
</font><font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_binary_hinge_'></a>loss_binary_hinge_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the hinge loss, which is
appropriate for binary classification problems. Therefore, the possible
labels when using this loss are +1 and -1. Moreover, it will cause the
network to produce outputs > 0 when predicting a member of the +1 class and
values < 0 otherwise.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> training_label_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the raw score for each classified object. If the score
is > 0 then the classifier is predicting the +1 class, otherwise it is
predicting the -1 class.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- all values pointed to by truth are +1 or -1.
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_binary_hinge <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_binary_hinge_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_binary_log_'></a>loss_binary_log_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the log loss, which is
appropriate for binary classification problems. Therefore, there are two possible
classes of labels: positive (> 0) and negative (< 0) when using this loss.
The absolute value of the label represents its weight. Putting a larger weight
on a sample increases the importance of getting its prediction correct during
training. A good rule of thumb is to use weights with absolute value 1 unless
you have a very unbalanced training dataset, in that case, give larger weight
to the class with less training examples.
This loss will cause the network to produce outputs > 0 when predicting a
member of the positive class and values < 0 otherwise.
To be more specific, this object contains a sigmoid layer followed by a
cross-entropy layer.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> training_label_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the raw score for each classified object. If the score
is > 0 then the classifier is predicting the +1 class, otherwise it is
predicting the -1 class.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- all values pointed to by truth are non-zero. Nominally they should be +1 or -1,
each indicating the desired class label.
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_binary_log <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_binary_log_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_multiclass_log_'></a>loss_multiclass_log_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the multiclass logistic
regression loss (e.g. negative log-likelihood loss), which is appropriate
for multiclass classification problems. This means that the possible
labels when using this loss are integers >= 0.
Moreover, if after training you were to replace the loss layer of the
network with a softmax layer, the network outputs would give the
probabilities of each class assignment. That is, if you have K classes
then the network should output tensors with the tensor::k()'th dimension
equal to K. Applying softmax to these K values gives the probabilities of
each class. The index into that K dimensional vector with the highest
probability is the predicted class label.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> training_label_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the predicted class for each classified object. The number
of possible output classes is sub.get_output().k().
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- all values pointed to by truth are < sub.get_output().k()
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_multiclass_log <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_multiclass_log_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> label_type<font color='#5555FF'>></font>
<font color='#0000FF'>struct</font> <b><a name='weighted_label'></a>weighted_label</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object represents the truth label of a single sample, together with
an associated weight (the higher the weight, the more emphasis the
corresponding sample is given during the training).
For technical reasons, it is defined in misc.h
This object is used in the following loss layers:
- loss_multiclass_log_weighted_ with unsigned long as label_type
- loss_multiclass_log_per_pixel_weighted_ with uint16_t as label_type,
since, in semantic segmentation, 65536 classes ought to be enough for
anybody.
!*/</font>
<b><a name='weighted_label'></a>weighted_label</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>
<b>{</b><b>}</b>
<b><a name='weighted_label'></a>weighted_label</b><font face='Lucida Console'>(</font>label_type label, <font color='#0000FF'><u>float</u></font> weight <font color='#5555FF'>=</font> <font color='#979000'>1.f</font><font face='Lucida Console'>)</font>
: label<font face='Lucida Console'>(</font>label<font face='Lucida Console'>)</font>, weight<font face='Lucida Console'>(</font>weight<font face='Lucida Console'>)</font>
<b>{</b><b>}</b>
<font color='#009900'>// The ground truth label
</font> label_type label<b>{</b><b>}</b>;
<font color='#009900'>// The weight of the corresponding sample
</font> <font color='#0000FF'><u>float</u></font> weight <font color='#5555FF'>=</font> <font color='#979000'>1.f</font>;
<b>}</b>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_multiclass_log_weighted_'></a>loss_multiclass_log_weighted_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the multiclass logistic
regression loss (e.g. negative log-likelihood loss), which is appropriate
for multiclass classification problems. It is basically just like the
loss_multiclass_log except that it lets you define per-sample weights,
which might be useful e.g. if you want to emphasize rare classes while
training. If the classification problem is difficult, a flat weight
structure may lead the network to always predict the most common label,
in particular if the degree of imbalance is high. To emphasize a certain
class or classes, simply increase the weights of the corresponding samples,
relative to the weights of other pixels.
Note that if you set all the weights equals to 1, then you get
loss_multiclass_log_ as a special case.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> dlib::weighted_label<font color='#5555FF'><</font><font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font><font color='#5555FF'>></font> weighted_label;
<font color='#0000FF'>typedef</font> weighted_label training_label_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the predicted class for each classified object. The number
of possible output classes is sub.get_output().k().
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- all values pointed to by truth are < sub.get_output().k()
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_multiclass_log_weighted <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_multiclass_log_weighted_, SUBNET<font color='#5555FF'>></font>;<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_multimulticlass_log_'></a>loss_multimulticlass_log_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements a collection of
multiclass classifiers. An example will make its use clear. So suppose,
for example, that you want to make something that takes a picture of a
vehicle and answers the following questions:
- What type of vehicle is it? A sedan or a truck?
- What color is it? red, green, blue, gray, or black?
You need two separate multi-class classifiers to do this. One to decide
the type of vehicle, and another to decide the color. The
loss_multimulticlass_log_ allows you to pack these two classifiers into one
neural network. This means that when you use the network to process an
image it will output 2 labels for each image, the type label and the color
label.
To create a loss_multimulticlass_log_ for the above case you would
construct it as follows:
std::map<std::string,std::vector<std::string>> labels;
labels["type"] = {"sedan", "truck"};
labels["color"] = {"red", "green", "blue", "gray", "black"};
loss_multimulticlass_log_ myloss(labels);
Then you could use myloss with a network object and train it to do this
task. More generally, you can use any number of classifiers and labels
when using this object. Finally, each of the classifiers uses a standard
multi-class logistic regression loss.
!*/</font>
<font color='#0000FF'>public</font>:
<b><a name='loss_multimulticlass_log_'></a>loss_multimulticlass_log_</b><font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- #number_of_labels() == 0
- #get_labels().size() == 0
!*/</font>
<b><a name='loss_multimulticlass_log_'></a>loss_multimulticlass_log_</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> std::map<font color='#5555FF'><</font>std::string,std::vector<font color='#5555FF'><</font>std::string<font color='#5555FF'>></font><font color='#5555FF'>></font><font color='#5555FF'>&</font> labels
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
requires
- Each vector in labels must contain at least 2 strings. I.e. each
classifier must have at least two possible labels.
ensures
- #number_of_labels() == the total number of strings in all the
std::vectors in labels.
- #number_of_classifiers() == labels.size()
- #get_labels() == labels
!*/</font>
<font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> <b><a name='number_of_labels'></a>number_of_labels</b><font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns the total number of labels known to this loss. This is the count of
all the labels in each classifier.
!*/</font>
<font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> <b><a name='number_of_classifiers'></a>number_of_classifiers</b><font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns the number of classifiers defined by this loss.
!*/</font>
std::map<font color='#5555FF'><</font>std::string,std::vector<font color='#5555FF'><</font>std::string<font color='#5555FF'>></font><font color='#5555FF'>></font> <b><a name='get_labels'></a>get_labels</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns the names of the classifiers and labels used by this loss. In
particular, if the returned object is L then:
- L[CLASS] == the set of labels used by the classifier CLASS.
- L.size() == number_of_classifiers()
- The count of strings in the vectors in L == number_of_labels()
!*/</font>
<font color='#0000FF'>class</font> <b><a name='classifier_output'></a>classifier_output</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object stores the predictions from one of the classifiers in
loss_multimulticlass_log_. It allows you to find out the most likely
string label predicted by that classifier, as well as get the class
conditional probability of any of the classes in the classifier.
!*/</font>
<font color='#0000FF'>public</font>:
<b><a name='classifier_output'></a>classifier_output</b><font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- #num_classes() == 0
!*/</font>
<font color='#0000FF'><u>size_t</u></font> <b><a name='num_classes'></a>num_classes</b><font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns the number of possible classes output by this classifier.
!*/</font>
<font color='#0000FF'><u>double</u></font> <b><a name='probability_of_class'></a>probability_of_class</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'><u>size_t</u></font> i
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
requires
- i < num_classes()
ensures
- returns the probability that the true class has a label of label(i).
- The sum of probability_of_class(j) for j in the range [0, num_classes()) is always 1.
!*/</font>
<font color='#0000FF'>const</font> std::string<font color='#5555FF'>&</font> <b><a name='label'></a>label</b><font face='Lucida Console'>(</font>
<font color='#0000FF'><u>size_t</u></font> i
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
requires
- i < num_classes()
ensures
- returns the string label for the ith class.
!*/</font>
<b><a name='operator'></a>operator</b> std::<b><a name='string'></a>string</b><font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
requires
- num_classes() != 0
ensures
- returns the string label for the most probable class.
!*/</font>
<font color='#0000FF'>friend</font> std::ostream<font color='#5555FF'>&</font> <b><a name='operator'></a>operator</b><font color='#5555FF'><</font><font color='#5555FF'><</font> <font face='Lucida Console'>(</font>std::ostream<font color='#5555FF'>&</font> out, <font color='#0000FF'>const</font> classifier_output<font color='#5555FF'>&</font> item<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
requires
- num_classes() != 0
ensures
- prints the most probable class label to out.
!*/</font>
<b>}</b>;
<font color='#009900'>// Both training_label_type and output_label_type should always have sizes equal to
</font> <font color='#009900'>// number_of_classifiers(). That is, the std::map should have an entry for every
</font> <font color='#009900'>// classifier known to this loss.
</font> <font color='#0000FF'>typedef</font> std::map<font color='#5555FF'><</font>std::string,std::string<font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> std::map<font color='#5555FF'><</font>std::string,classifier_output<font color='#5555FF'>></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- number_of_labels() != 0
- sub.get_output().k() == number_of_labels()
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- number_of_labels() != 0
- sub.get_output().k() == number_of_labels()
It should be noted that the last layer in your network should usually
be an fc layer. If so, you can satisfy this requirement of k() being
number_of_labels() by calling set_num_outputs() prior to training your
network like so:
your_network.subnet().layer_details().set_num_outputs(your_network.loss_details().number_of_labels());
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- All the std::maps pointed to by truth contain entries for all the
classifiers known to this loss. That is, it must be valid to call
truth[i][classifier] for any of the classifiers known to this loss. To
say this another way, all the training samples must contain labels for
each of the classifiers defined by this loss.
To really belabor this, this also means that truth[i].size() ==
get_labels().size() and that both truth[i] and get_labels() have the same
set of key strings. It also means that the value strings in truth[i]
must be strings known to the loss, i.e. they are valid labels according
to get_labels().
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_multimulticlass_log <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_multimulticlass_log_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// Allow comparison between classifier_outputs and std::string to check if the
</font> <font color='#009900'>// predicted class is a particular string.
</font> <font color='#0000FF'>inline</font> <font color='#0000FF'><u>bool</u></font> <b><a name='operator'></a>operator</b><font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font face='Lucida Console'>(</font><font color='#0000FF'>const</font> std::string<font color='#5555FF'>&</font> lhs, <font color='#0000FF'>const</font> loss_multimulticlass_log_::classifier_output<font color='#5555FF'>&</font> rhs<font face='Lucida Console'>)</font>
<b>{</b> <font color='#0000FF'>return</font> lhs <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#0000FF'>static_cast</font><font color='#5555FF'><</font><font color='#0000FF'>const</font> std::string<font color='#5555FF'>&</font><font color='#5555FF'>></font><font face='Lucida Console'>(</font>rhs<font face='Lucida Console'>)</font>; <b>}</b>
<font color='#0000FF'>inline</font> <font color='#0000FF'><u>bool</u></font> <b><a name='operator'></a>operator</b><font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font face='Lucida Console'>(</font><font color='#0000FF'>const</font> loss_multimulticlass_log_::classifier_output<font color='#5555FF'>&</font> lhs, <font color='#0000FF'>const</font> std::string<font color='#5555FF'>&</font> rhs<font face='Lucida Console'>)</font>
<b>{</b> <font color='#0000FF'>return</font> rhs <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#0000FF'>static_cast</font><font color='#5555FF'><</font><font color='#0000FF'>const</font> std::string<font color='#5555FF'>&</font><font color='#5555FF'>></font><font face='Lucida Console'>(</font>lhs<font face='Lucida Console'>)</font>; <b>}</b>
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_multibinary_log_'></a>loss_multibinary_log_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements a collection of
binary classifiers using the log loss, which is appropriate for
binary classification problems where each sample can belong to zero
or more categories. Therefore, there are two possible classes of labels:
positive (> 0) and negative (< 0) when using this loss.
The absolute value of the label represents its weight. Putting a larger
weight on a sample increases its importance of getting its prediction
correct during training. A good rule of thumb is to use weights with
absolute value 1 unless you have a very unbalanced training dataset,
in that case, give larger weight to the class with less training examples.
This loss will cause the network to produce outputs > 0 when predicting a
member of the positive classes and values < 0 otherwise.
To be more specific, this object contains a sigmoid layer followed by a
cross-entropy layer.
An example will make its use clear. So suppose, for example, that you want
to make a classifier for cats and dogs, but what happens if they both
appear in one image? Or none of them? This layer allows you to handle
those use cases by using the following labels:
- std::vector<float> dog_label = {1.f, -1.f};
- std::vector<float> cat_label = {-1.f , 1.f};
- std::vector<float> both_label = {1.f, 1.f};
- std::vector<float> none_label = {-1.f, -1.f};
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> std::vector<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> std::vector<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output labels are the raw scores for each classified object. If a score
is > 0 then the classifier is predicting the +1 class for that category, otherwise
it is predicting the -1 class.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- truth points to training_label_type elements, each of size sub.get_output.k().
The elements of each truth training_label_type instance are nominally +1 or -1,
each representing a binary class label.
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_multibinary_log <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_multibinary_log_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font><font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>enum</font> <font color='#0000FF'>class</font> <b><a name='use_image_pyramid'></a>use_image_pyramid</b> : uint8_t
<b>{</b>
no,
yes
<b>}</b>;
<font color='#0000FF'>struct</font> <b><a name='mmod_options'></a>mmod_options</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object contains all the parameters that control the behavior of loss_mmod_.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>struct</font> <b><a name='detector_window_details'></a>detector_window_details</b>
<b>{</b>
<b><a name='detector_window_details'></a>detector_window_details</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>=</font> <font color='#0000FF'>default</font>;
<b><a name='detector_window_details'></a>detector_window_details</b><font face='Lucida Console'>(</font><font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> w, <font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> h<font face='Lucida Console'>)</font> : width<font face='Lucida Console'>(</font>w<font face='Lucida Console'>)</font>, height<font face='Lucida Console'>(</font>h<font face='Lucida Console'>)</font> <b>{</b><b>}</b>
<b><a name='detector_window_details'></a>detector_window_details</b><font face='Lucida Console'>(</font><font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> w, <font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> h, <font color='#0000FF'>const</font> std::string<font color='#5555FF'>&</font> l<font face='Lucida Console'>)</font> : width<font face='Lucida Console'>(</font>w<font face='Lucida Console'>)</font>, height<font face='Lucida Console'>(</font>h<font face='Lucida Console'>)</font>, label<font face='Lucida Console'>(</font>l<font face='Lucida Console'>)</font> <b>{</b><b>}</b>
<font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> width <font color='#5555FF'>=</font> <font color='#979000'>0</font>;
<font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> height <font color='#5555FF'>=</font> <font color='#979000'>0</font>;
std::string label;
<font color='#0000FF'>friend</font> <font color='#0000FF'>inline</font> <font color='#0000FF'><u>void</u></font> <b><a name='serialize'></a>serialize</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> detector_window_details<font color='#5555FF'>&</font> item, std::ostream<font color='#5555FF'>&</font> out<font face='Lucida Console'>)</font>;
<font color='#0000FF'>friend</font> <font color='#0000FF'>inline</font> <font color='#0000FF'><u>void</u></font> <b><a name='deserialize'></a>deserialize</b><font face='Lucida Console'>(</font>detector_window_details<font color='#5555FF'>&</font> item, std::istream<font color='#5555FF'>&</font> in<font face='Lucida Console'>)</font>;
<b>}</b>;
<b><a name='mmod_options'></a>mmod_options</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>=</font> <font color='#0000FF'>default</font>;
<font color='#009900'>// This kind of object detector is a sliding window detector. The detector_windows
</font> <font color='#009900'>// field determines how many sliding windows we will use and what the shape of each
</font> <font color='#009900'>// window is. It also determines the output label applied to each detection
</font> <font color='#009900'>// identified by each window. Since you will usually use the MMOD loss with an
</font> <font color='#009900'>// image pyramid, the detector sizes also determine the size of the smallest object
</font> <font color='#009900'>// you can detect.
</font> std::vector<font color='#5555FF'><</font>detector_window_details<font color='#5555FF'>></font> detector_windows;
<font color='#009900'>// These parameters control how we penalize different kinds of mistakes. See
</font> <font color='#009900'>// Max-Margin Object Detection by Davis E. King (http://arxiv.org/abs/1502.00046)
</font> <font color='#009900'>// for further details.
</font> <font color='#0000FF'><u>double</u></font> loss_per_false_alarm <font color='#5555FF'>=</font> <font color='#979000'>1</font>;
<font color='#0000FF'><u>double</u></font> loss_per_missed_target <font color='#5555FF'>=</font> <font color='#979000'>1</font>;
<font color='#009900'>// A detection must have an intersection-over-union value greater than this for us
</font> <font color='#009900'>// to consider it a match against a ground truth box.
</font> <font color='#0000FF'><u>double</u></font> truth_match_iou_threshold <font color='#5555FF'>=</font> <font color='#979000'>0.5</font>;
<font color='#009900'>// When doing non-max suppression, we use overlaps_nms to decide if a box overlaps
</font> <font color='#009900'>// an already output detection and should therefore be thrown out.
</font> test_box_overlap overlaps_nms <font color='#5555FF'>=</font> <b><a name='test_box_overlap'></a>test_box_overlap</b><font face='Lucida Console'>(</font><font color='#979000'>0.4</font><font face='Lucida Console'>)</font>;
<font color='#009900'>// Any mmod_rect in the training data that has its ignore field set to true defines
</font> <font color='#009900'>// an "ignore zone" in an image. Any detection from that area is totally ignored
</font> <font color='#009900'>// by the optimizer. Therefore, this overlaps_ignore field defines how we decide
</font> <font color='#009900'>// if a box falls into an ignore zone. You use these ignore zones if there are
</font> <font color='#009900'>// objects in your dataset that you are unsure if you want to detect or otherwise
</font> <font color='#009900'>// don't care if the detector gets them or not.
</font> test_box_overlap overlaps_ignore;
<font color='#009900'>// Usually the detector would be scale-invariant, and used with an image pyramid.
</font> <font color='#009900'>// However, sometimes scale-invariance may not be desired.
</font> use_image_pyramid assume_image_pyramid <font color='#5555FF'>=</font> use_image_pyramid::yes;
<font color='#009900'>// By default, the mmod loss doesn't train any bounding box regression model. But
</font> <font color='#009900'>// if you set use_bounding_box_regression == true then it expects the network to
</font> <font color='#009900'>// output a tensor with detector_windows.size()*5 channels rather than just
</font> <font color='#009900'>// detector_windows.size() channels. The 4 extra channels per window are trained
</font> <font color='#009900'>// to give a bounding box regression output that improves the positioning of the
</font> <font color='#009900'>// output detection box.
</font> <font color='#0000FF'><u>bool</u></font> use_bounding_box_regression <font color='#5555FF'>=</font> <font color='#979000'>false</font>;
<font color='#009900'>// When using bounding box regression, bbr_lambda determines how much you care
</font> <font color='#009900'>// about getting the bounding box shape correct vs just getting the detector to
</font> <font color='#009900'>// find objects. That is, the objective function being optimized is
</font> <font color='#009900'>// basic_mmod_loss + bbr_lambda*bounding_box_regression_loss. So setting
</font> <font color='#009900'>// bbr_lambda to a larger value will cause the overall loss to care more about
</font> <font color='#009900'>// getting the bounding box shape correct.
</font> <font color='#0000FF'><u>double</u></font> bbr_lambda <font color='#5555FF'>=</font> <font color='#979000'>100</font>;
<font color='#009900'>// Tell the loss not to print warnings about impossible labels. You should think very hard
</font> <font color='#009900'>// before turning this off as it's very often telling you something is really wrong with
</font> <font color='#009900'>// your training data.
</font> <font color='#0000FF'><u>bool</u></font> be_quiet <font color='#5555FF'>=</font> <font color='#979000'>false</font>;
<b><a name='mmod_options'></a>mmod_options</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> std::vector<font color='#5555FF'><</font>std::vector<font color='#5555FF'><</font>mmod_rect<font color='#5555FF'>></font><font color='#5555FF'>></font><font color='#5555FF'>&</font> boxes,
<font color='#0000FF'>const</font> <font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> target_size,
<font color='#0000FF'>const</font> <font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> min_target_size,
<font color='#0000FF'>const</font> <font color='#0000FF'><u>double</u></font> min_detector_window_overlap_iou <font color='#5555FF'>=</font> <font color='#979000'>0.75</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
requires
- 0 < min_target_size <= target_size
- 0.5 < min_detector_window_overlap_iou < 1
ensures
- use_image_pyramid_ == use_image_pyramid::yes
- This function should be used when scale-invariance is desired, and
input_rgb_image_pyramid is therefore used as the input layer.
- This function tries to automatically set the MMOD options to reasonable
values, assuming you have a training dataset of boxes.size() images, where
the ith image contains objects boxes[i] you want to detect.
- The most important thing this function does is decide what detector
windows should be used. This is done by finding a set of detector
windows that are sized such that:
- When slid over an image pyramid, each box in boxes will have an
intersection-over-union with one of the detector windows of at least
min_detector_window_overlap_iou. That is, we will make sure that
each box in boxes could potentially be detected by one of the
detector windows. This essentially comes down to picking detector
windows with aspect ratios similar to the aspect ratios in boxes.
Note that we also make sure that each box can be detected by a window
with the same label. For example, if all the boxes had the same
aspect ratio but there were 4 different labels used in boxes then
there would be 4 resulting detector windows, one for each label.
- The longest edge of each detector window is target_size pixels in
length, unless the window's shortest side would be less than
min_target_size pixels in length. In this case the shortest side
will be set to min_target_size length, and the other side sized to
preserve the aspect ratio of the window.
This means that target_size and min_target_size control the size of the
detector windows, while the aspect ratios of the detector windows are
automatically determined by the contents of boxes. It should also be
emphasized that the detector isn't going to be able to detect objects
smaller than any of the detector windows. So consider that when setting
these sizes.
- This function will also set the overlaps_nms tester to the most
restrictive tester that doesn't reject anything in boxes.
!*/</font>
<b><a name='mmod_options'></a>mmod_options</b> <font face='Lucida Console'>(</font>
use_image_pyramid use_image_pyramid,
<font color='#0000FF'>const</font> std::vector<font color='#5555FF'><</font>std::vector<font color='#5555FF'><</font>mmod_rect<font color='#5555FF'>></font><font color='#5555FF'>></font><font color='#5555FF'>&</font> boxes,
<font color='#0000FF'>const</font> <font color='#0000FF'><u>double</u></font> min_detector_window_overlap_iou <font color='#5555FF'>=</font> <font color='#979000'>0.75</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
requires
- use_image_pyramid == use_image_pyramid::no
- 0.5 < min_detector_window_overlap_iou < 1
ensures
- This function should be used when scale-invariance is not desired, and
there is no intention to apply an image pyramid.
- This function tries to automatically set the MMOD options to reasonable
values, assuming you have a training dataset of boxes.size() images, where
the ith image contains objects boxes[i] you want to detect.
- The most important thing this function does is decide what detector
windows should be used. This is done by finding a set of detector
windows that are sized such that:
- When slid over an image, each box in boxes will have an
intersection-over-union with one of the detector windows of at least
min_detector_window_overlap_iou. That is, we will make sure that
each box in boxes could potentially be detected by one of the
detector windows.
- This function will also set the overlaps_nms tester to the most
restrictive tester that doesn't reject anything in boxes.
!*/</font>
<b>}</b>;
<font color='#0000FF'><u>void</u></font> <b><a name='serialize'></a>serialize</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> mmod_options<font color='#5555FF'>&</font> item, std::ostream<font color='#5555FF'>&</font> out<font face='Lucida Console'>)</font>;
<font color='#0000FF'><u>void</u></font> <b><a name='deserialize'></a>deserialize</b><font face='Lucida Console'>(</font>mmod_options<font color='#5555FF'>&</font> item, std::istream<font color='#5555FF'>&</font> in<font face='Lucida Console'>)</font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_mmod_'></a>loss_mmod_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the Max Margin Object
Detection loss defined in the paper:
Max-Margin Object Detection by Davis E. King (http://arxiv.org/abs/1502.00046).
This means you use this loss if you want to detect the locations of objects
in images.
It should also be noted that this loss layer requires an input layer that
defines the following functions:
- image_contained_point()
- tensor_space_to_image_space()
- image_space_to_tensor_space()
A reference implementation of them and their definitions can be found in
the input_rgb_image_pyramid object, which is the recommended input layer to
be used with loss_mmod_.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> std::vector<font color='#5555FF'><</font>mmod_rect<font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> std::vector<font color='#5555FF'><</font>mmod_rect<font color='#5555FF'>></font> output_label_type;
<b><a name='loss_mmod_'></a>loss_mmod_</b><font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- #get_options() == mmod_options()
!*/</font>
<b><a name='loss_mmod_'></a>loss_mmod_</b><font face='Lucida Console'>(</font>
mmod_options options_
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- #get_options() == options_
!*/</font>
<font color='#0000FF'>const</font> mmod_options<font color='#5555FF'>&</font> <b><a name='get_options'></a>get_options</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns the options object that defines the general behavior of this loss layer.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter,
<font color='#0000FF'><u>double</u></font> adjust_threshold <font color='#5555FF'>=</font> <font color='#979000'>0</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
Also, the output labels are std::vectors of mmod_rects where, for each mmod_rect R,
we have the following interpretations:
- R.rect == the location of an object in the image.
- R.detection_confidence the score for the object, the bigger the score the
more confident the detector is that an object is really there. Only
objects with a detection_confidence > adjust_threshold are output. So if
you want to output more objects (that are also of less confidence) you
can call to_label() with a smaller value of adjust_threshold.
- R.ignore == false (this value is unused by to_label()).
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
Also, the loss value returned is roughly equal to the average number of
mistakes made per image. This is the sum of false alarms and missed
detections, weighted by the loss weights for these types of mistakes specified
in the mmod_options.
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_mmod <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_mmod_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_metric_'></a>loss_metric_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it allows you to learn to map objects
into a vector space where objects sharing the same class label are close to
each other, while objects with different labels are far apart.
To be specific, it optimizes the following loss function which considers
all pairs of objects in a mini-batch and computes a different loss depending
on their respective class labels. So if objects A1 and A2 in a mini-batch
share the same class label then their contribution to the loss is:
max(0, length(A1-A2)-get_distance_threshold() + get_margin())
While if A1 and B1 have different class labels then their contribution to
the loss function is:
max(0, get_distance_threshold()-length(A1-B1) + get_margin())
Therefore, this loss layer optimizes a version of the hinge loss.
Moreover, the loss is trying to make sure that all objects with the same
label are within get_distance_threshold() distance of each other.
Conversely, if two objects have different labels then they should be more
than get_distance_threshold() distance from each other in the learned
embedding. So this loss function gives you a natural decision boundary for
deciding if two objects are from the same class.
Finally, the loss balances the number of negative pairs relative to the
number of positive pairs. Therefore, if there are N pairs that share the
same identity in a mini-batch then the algorithm will only include the N
worst non-matching pairs in the loss. That is, the algorithm performs hard
negative mining on the non-matching pairs. This is important since there
are in general way more non-matching pairs than matching pairs. So to
avoid imbalance in the loss this kind of hard negative mining is useful.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>unsigned</u></font> <font color='#0000FF'><u>long</u></font> training_label_type;
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font>,<font color='#979000'>0</font>,<font color='#979000'>1</font><font color='#5555FF'>></font> output_label_type;
<b><a name='loss_metric_'></a>loss_metric_</b><font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- #get_margin() == 0.04
- #get_distance_threshold() == 0.6
!*/</font>
<b><a name='loss_metric_'></a>loss_metric_</b><font face='Lucida Console'>(</font>
<font color='#0000FF'><u>float</u></font> margin,
<font color='#0000FF'><u>float</u></font> dist_thresh
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
requires
- margin > 0
- dist_thresh > 0
ensures
- #get_margin() == margin
- #get_distance_threshold() == dist_thresh
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
This loss expects the network to produce a single vector (per sample) as
output. This vector is the learned embedding. Therefore, to_label() just
copies these output vectors from the network into the output label_iterators
given to this function, one for each sample in the input_tensor.
!*/</font>
<font color='#0000FF'><u>float</u></font> <b><a name='get_margin'></a>get_margin</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns the margin value used by the loss function. See the discussion
in WHAT THIS OBJECT REPRESENTS for details.
!*/</font>
<font color='#0000FF'><u>float</u></font> <b><a name='get_distance_threshold'></a>get_distance_threshold</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns the distance threshold value used by the loss function. See the discussion
in WHAT THIS OBJECT REPRESENTS for details.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_metric <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_metric_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_ranking_'></a>loss_ranking_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the pairwise ranking
loss described in the paper:
Optimizing Search Engines using Clickthrough Data by Thorsten Joachims
This is the same loss function used by the dlib::svm_rank_trainer object.
Therefore, it is generally appropriate when you have a two class problem
and you want to learn a function that ranks one class before the other.
So for example, suppose you have two classes of data. Objects of type A
and objects of type B. Moreover, suppose that you want to sort the objects
so that A objects always come before B objects. This loss will help you
learn a function that assigns a real number to each object such that A
objects get a larger number assigned to them than B objects. This lets you
then sort the objects according to the output of the neural network and
obtain the desired result of having A objects come before B objects.
The training labels should be positive values for objects you want to get
high scores and negative for objects that should get small scores. So
relative to our A/B example, you would give A objects labels of +1 and B
objects labels of -1. This should cause the learned network to give A
objects large positive values and B objects negative values.
Finally, the specific loss function is:
For all pairs of positive vs negative training examples A_i and B_j respectively:
sum_ij: max(0, B_i - A_j + margin_ij)
where margin_ij = the label for A_j minus the label for B_i. If you
always use +1 and -1 labels then the margin is always 2. However, this
formulation allows you to give certain training samples different weight by
adjusting the training labels appropriately.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> training_label_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the predicted ranking score.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_ranking <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_ranking_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_epsilon_insensitive_'></a>loss_epsilon_insensitive_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the epsilon insensitive
loss, which is appropriate for regression problems. In particular, this
loss function is;
loss(y1,y2) = abs(y1-y2)<epsilon ? 0 : abs(y1-y2)-epsilon
Therefore, the loss is basically just the abs() loss except there is a dead
zone around zero, causing the loss to not care about mistakes of magnitude
smaller than epsilon.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> training_label_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> output_label_type;
<b><a name='loss_epsilon_insensitive_'></a>loss_epsilon_insensitive_</b><font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#5555FF'>=</font> <font color='#0000FF'>default</font>;
<font color='#009900'>/*!
ensures
- #get_epsilon() == 1
!*/</font>
<b><a name='loss_epsilon_insensitive_'></a>loss_epsilon_insensitive_</b><font face='Lucida Console'>(</font>
<font color='#0000FF'><u>double</u></font> eps
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
requires
- eps >= 0
ensures
- #get_epsilon() == eps
!*/</font>
<font color='#0000FF'><u>double</u></font> <b><a name='get_epsilon'></a>get_epsilon</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns the epsilon value used in the loss function. Mistakes in the
regressor smaller than get_epsilon() are ignored by the loss function.
!*/</font>
<font color='#0000FF'><u>void</u></font> <b><a name='set_epsilon'></a>set_epsilon</b><font face='Lucida Console'>(</font>
<font color='#0000FF'><u>double</u></font> eps
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
requires
- eps >= 0
ensures
- #get_epsilon() == eps
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the predicted continuous variable.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_epsilon_insensitive <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_epsilon_insensitive_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_mean_squared_'></a>loss_mean_squared_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the mean squared loss, which is
appropriate for regression problems.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> training_label_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'><u>float</u></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the predicted continuous variable.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_mean_squared <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_mean_squared_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_mean_squared_multioutput_'></a>loss_mean_squared_multioutput_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the mean squared loss,
which is appropriate for regression problems. It is basically just like
loss_mean_squared_ except that it lets you define multiple outputs instead
of just 1.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the predicted continuous variable.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().nr() == 1
- sub.get_output().nc() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- (*(truth + idx)).nc() == 1 for all idx such that 0 <= idx < sub.get_output().num_samples()
- (*(truth + idx)).nr() == sub.get_output().k() for all idx such that 0 <= idx < sub.get_output().num_samples()
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_mean_squared_multioutput <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_mean_squared_multioutput_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_binary_log_per_pixel_'></a>loss_binary_log_per_pixel_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the log loss, which is
appropriate for binary classification problems. It is basically just like
loss_binary_log_ except that it lets you define matrix outputs instead
of scalar outputs. It should be useful, for example, in segmentation
where we want to classify each pixel of an image, and also get at least
some sort of confidence estimate for each pixel.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the raw score for each classified object. If the score
is > 0 then the classifier is predicting the +1 class, otherwise it is
predicting the -1 class.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- all pixel values pointed to by truth correspond to the desired target values.
Nominally they should be +1 or -1, each indicating the desired class label,
or 0 to indicate that the corresponding pixel is to be ignored.
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_binary_log_per_pixel <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_binary_log_per_pixel_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_multiclass_log_per_pixel_'></a>loss_multiclass_log_per_pixel_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the multiclass logistic
regression loss (e.g. negative log-likelihood loss), which is appropriate
for multiclass classification problems. It is basically just like
loss_multiclass_log_ except that it lets you define matrix outputs instead
of scalar outputs. It should be useful, for example, in semantic
segmentation where we want to classify each pixel of an image.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#009900'>// In semantic segmentation, if you don't know the ground-truth of some pixel,
</font> <font color='#009900'>// set the label of that pixel to this value. When you do so, the pixel will be
</font> <font color='#009900'>// ignored when computing gradients.
</font> <font color='#0000FF'>static</font> <font color='#0000FF'>const</font> uint16_t label_to_ignore <font color='#5555FF'>=</font> std::numeric_limits<font color='#5555FF'><</font>uint16_t<font color='#5555FF'>></font>::<b><a name='max'></a>max</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>;
<font color='#009900'>// In semantic segmentation, 65535 classes ought to be enough for anybody.
</font> <font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font>uint16_t<font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font>uint16_t<font color='#5555FF'>></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the predicted class for each classified element. The number
of possible output classes is sub.get_output().k().
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- all values pointed to by truth are < sub.get_output().k() or are equal to label_to_ignore.
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_multiclass_log_per_pixel <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_multiclass_log_per_pixel_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_multiclass_log_per_pixel_weighted_'></a>loss_multiclass_log_per_pixel_weighted_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the multiclass logistic
regression loss (e.g. negative log-likelihood loss), which is appropriate
for multiclass classification problems. It is basically just like
loss_multiclass_log_per_pixel_ except that it lets you define per-pixel
weights, which may be useful e.g. if you want to emphasize rare classes
while training. (If the classification problem is difficult, a flat weight
structure may lead the network to always predict the most common label, in
particular if the degree of imbalance is high. To emphasize a certain
class or classes, simply increase the weights of the corresponding pixels,
relative to the weights of the other pixels.)
Note that if you set the weight to 0 whenever a pixel's label is equal to
loss_multiclass_log_per_pixel_::label_to_ignore, and to 1 otherwise, then
you essentially get loss_multiclass_log_per_pixel_ as a special case.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> dlib::weighted_label<font color='#5555FF'><</font>uint16_t<font color='#5555FF'>></font> weighted_label;
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font>weighted_label<font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font>uint16_t<font color='#5555FF'>></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output label is the predicted class for each classified element. The number
of possible output classes is sub.get_output().k().
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- all labels pointed to by truth are < sub.get_output().k(), or the corresponding weight
is zero.
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_multiclass_log_per_pixel_weighted <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_multiclass_log_per_pixel_weighted_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_mean_squared_per_pixel_'></a>loss_mean_squared_per_pixel_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the mean squared loss,
which is appropriate for regression problems. It is basically just like
loss_mean_squared_multioutput_ except that it lets you define matrix or
image outputs, instead of vector.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output labels are the predicted continuous variables.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().k() == 1
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- for all idx such that 0 <= idx < sub.get_output().num_samples():
- sub.get_output().nr() == (*(truth + idx)).nr()
- sub.get_output().nc() == (*(truth + idx)).nc()
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_mean_squared_per_pixel <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_mean_squared_per_pixel_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>template</font><font color='#5555FF'><</font><font color='#0000FF'><u>long</u></font> _num_channels<font color='#5555FF'>></font>
<font color='#0000FF'>class</font> <b><a name='loss_mean_squared_per_channel_and_pixel_'></a>loss_mean_squared_per_channel_and_pixel_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, it implements the mean squared loss,
which is appropriate for regression problems. It is basically just like
loss_mean_squared_per_pixel_ except that it computes the loss over all
channels, not just the first one.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> std::array<font color='#5555FF'><</font>matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font>, _num_channels<font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> std::array<font color='#5555FF'><</font>matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>></font>, _num_channels<font color='#5555FF'>></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.get_output().k() == _num_channels
- sub.sample_expansion_factor() == 1
and the output labels are the predicted continuous variables.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().k() == _num_channels
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- for all idx such that 0 <= idx < sub.get_output().num_samples():
- sub.get_output().nr() == (*(truth + idx)).nr()
- sub.get_output().nc() == (*(truth + idx)).nc()
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'><u>long</u></font> num_channels, <font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_mean_squared_per_channel_and_pixel <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_mean_squared_per_channel_and_pixel_<font color='#5555FF'><</font>num_channels<font color='#5555FF'>></font>, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<font color='#0000FF'>class</font> <b><a name='loss_dot_'></a>loss_dot_</b>
<b>{</b>
<font color='#009900'>/*!
WHAT THIS OBJECT REPRESENTS
This object implements the loss layer interface defined above by
EXAMPLE_LOSS_LAYER_. In particular, selecting this loss means you want
maximize the dot product between the output of a network and a set of
training vectors. The loss is therefore the negative dot product. To be
very specific, if X is the output vector of a network and Y is a training
label (also a vector), then the loss for this training sample is: -dot(X,Y)
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font>,<font color='#979000'>0</font>,<font color='#979000'>1</font><font color='#5555FF'>></font> training_label_type;
<font color='#0000FF'>typedef</font> matrix<font color='#5555FF'><</font><font color='#0000FF'><u>float</u></font>,<font color='#979000'>0</font>,<font color='#979000'>1</font><font color='#5555FF'>></font> output_label_type;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> SUB_TYPE,
<font color='#0000FF'>typename</font> label_iterator
<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='to_label'></a>to_label</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
<font color='#0000FF'>const</font> SUB_TYPE<font color='#5555FF'>&</font> sub,
label_iterator iter
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::to_label() except
it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
and the output labels are simply the final network outputs stuffed into a
vector. To be very specific, the output is the following for all valid i:
*(iter+i) == trans(rowm(mat(sub.get_output()),i))
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> const_label_iterator,
<font color='#0000FF'>typename</font> SUBNET
<font color='#5555FF'>></font>
<font color='#0000FF'><u>double</u></font> <b><a name='compute_loss_value_and_gradient'></a>compute_loss_value_and_gradient</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> input_tensor,
const_label_iterator truth,
SUBNET<font color='#5555FF'>&</font> sub
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
This function has the same interface as EXAMPLE_LOSS_LAYER_::compute_loss_value_and_gradient()
except it has the additional calling requirements that:
- sub.get_output().num_samples() == input_tensor.num_samples()
- sub.sample_expansion_factor() == 1
- Let NETWORK_OUTPUT_DIMS == sub.get_output().size()/sub.get_output().num_samples()
- for all idx such that 0 <= idx < sub.get_output().num_samples():
- NETWORK_OUTPUT_DIMS == (*(truth + idx)).size()
!*/</font>
<b>}</b>;
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> SUBNET<font color='#5555FF'>></font>
<font color='#0000FF'>using</font> loss_dot <font color='#5555FF'>=</font> add_loss_layer<font color='#5555FF'><</font>loss_dot_, SUBNET<font color='#5555FF'>></font>;
<font color='#009900'>// ----------------------------------------------------------------------------------------
</font>
<b>}</b>
<font color='#0000FF'>#endif</font> <font color='#009900'>// DLIB_DNn_LOSS_ABSTRACT_H_
</font>
</pre></body></html> |