drone-DinoVdeau-produttoria_binary-probabilities is a fine-tuned version of drone-DinoVdeau-produttoria_binary-probabilities-large-2024_11_03-batch-size64_freeze_probs. It achieves the following results on the test set:

Loss: 0.3499
F1 Micro: 0.8642
F1 Macro: 0.7488
Accuracy: 0.1948
RMSE: 0.1848
MAE: 0.1248
R2: 0.4361

Class	F1 per class
Acropore_branched	0.7745
Acropore_digitised	0.5393
Acropore_tabular	0.4310
Algae	0.9823
Dead_coral	0.7850
Fish	0.6550
Millepore	0.4152
No_acropore_encrusting	0.6755
No_acropore_massive	0.7954
No_acropore_sub_massive	0.7421
Rock	0.9888
Rubble	0.9645
Sand	0.9853

Model description

drone-DinoVdeau-produttoria_binary-probabilities is a model built on top of drone-DinoVdeau-produttoria_binary-probabilities-large-2024_11_03-batch-size64_freeze_probs model for underwater multilabel image classification.The classification head is a combination of linear, ReLU, batch normalization, and dropout layers.

The source code for training the model can be found in this Git repository.

Developed by: lombardata, credits to César Leblanc and Victor Illien

Intended uses & limitations

You can use the raw model for classify diverse marine species, encompassing coral morphotypes classes taken from the Global Coral Reef Monitoring Network (GCRMN), habitats classes and seagrass species.

Training and evaluation data

Details on the estimated number of images for each class are given in the following table:

Class	train	test	val	Total
Acropore_branched	1282	458	480	2220
Acropore_digitised	800	276	253	1329
Acropore_tabular	377	133	135	645
Algae	9284	3133	3167	15584
Dead_coral	3207	1097	1103	5407
Fish	1360	481	487	2328
Millepore	258	110	94	462
No_acropore_encrusting	981	403	407	1791
No_acropore_massive	3261	1225	1291	5777
No_acropore_sub_massive	2371	857	889	4117
Rock	10077	3388	3402	16867
Rubble	8590	2880	2878	14348
Sand	9880	3283	3311	16474

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

Number of Epochs: 68.0
Learning Rate: 0.001
Train Batch Size: 64
Eval Batch Size: 64
Optimizer: Adam
LR Scheduler Type: ReduceLROnPlateau with a patience of 5 epochs and a factor of 0.1
Freeze Encoder: Yes
Data Augmentation: Yes

Data Augmentation

Data were augmented using the following transformations :

Train Transforms

PreProcess: No additional parameters
Resize: probability=1.00
RandomHorizontalFlip: probability=0.25
RandomVerticalFlip: probability=0.25
ColorJiggle: probability=0.25
RandomPerspective: probability=0.25
Normalize: probability=1.00

Val Transforms

PreProcess: No additional parameters
Resize: probability=1.00
Normalize: probability=1.00

Training results

Epoch	Validation Loss	MAE	RMSE	R2	Learning Rate
1	0.3794757127761841	0.14892756938934326	0.20668388903141022	0.2894136607646942	0.001
2	0.3673744201660156	0.1374034583568573	0.198309525847435	0.35173478722572327	0.001
3	0.3671453297138214	0.1413687765598297	0.198079913854599	0.35207054018974304	0.001
4	0.36317145824432373	0.1391323208808899	0.19521364569664001	0.3708474040031433	0.001
5	0.3678734302520752	0.141828715801239	0.19933271408081055	0.3453221321105957	0.001
6	0.36250030994415283	0.13803647458553314	0.19508354365825653	0.37177884578704834	0.001
7	0.36188462376594543	0.1347939670085907	0.19410446286201477	0.3771490454673767	0.001
8	0.36125126481056213	0.13681310415267944	0.19354337453842163	0.37883251905441284	0.001
9	0.3603876233100891	0.135352224111557	0.19339486956596375	0.3812035918235779	0.001
10	0.3612792491912842	0.13375206291675568	0.19321060180664062	0.38122493028640747	0.001
11	0.3603772521018982	0.13226205110549927	0.19312407076358795	0.3844555616378784	0.001
12	0.361823707818985	0.13859649002552032	0.1941623091697693	0.3774065375328064	0.001
13	0.35931822657585144	0.13433586061000824	0.1924724578857422	0.3875495195388794	0.001
14	0.3604746460914612	0.13521355390548706	0.19319292902946472	0.38306838274002075	0.001
15	0.36050480604171753	0.13660094141960144	0.19354429841041565	0.3816676735877991	0.001
16	0.3599933683872223	0.13121920824050903	0.1921611875295639	0.388213574886322	0.001
17	0.3628774583339691	0.13783428072929382	0.1932491511106491	0.38425371050834656	0.001
18	0.36154037714004517	0.1323489546775818	0.19430074095726013	0.37684857845306396	0.001
19	0.3594801127910614	0.13296250998973846	0.19218452274799347	0.38950252532958984	0.001
20	0.3565874397754669	0.13302744925022125	0.19017010927200317	0.40064936876296997	0.0001
21	0.35486647486686707	0.13062793016433716	0.18895885348320007	0.40758493542671204	0.0001
22	0.35447388887405396	0.13081100583076477	0.18863680958747864	0.4096067547798157	0.0001
23	0.3544616997241974	0.13033078610897064	0.18823565542697906	0.411631315946579	0.0001
24	0.3539991080760956	0.1316699981689453	0.1881898045539856	0.4120980501174927	0.0001
25	0.3545873463153839	0.12844440340995789	0.18831981718540192	0.41126883029937744	0.0001
26	0.3529074192047119	0.1263934224843979	0.18757320940494537	0.4154190421104431	0.0001
27	0.3532767593860626	0.129387766122818	0.187411367893219	0.41658732295036316	0.0001
28	0.3532498776912689	0.12938687205314636	0.18755248188972473	0.41600102186203003	0.0001
29	0.35306474566459656	0.1302015781402588	0.1871432662010193	0.41835859417915344	0.0001
30	0.3536038398742676	0.1291646808385849	0.18775980174541473	0.414754718542099	0.0001
31	0.3530591130256653	0.1267225444316864	0.1876552253961563	0.415239542722702	0.0001
32	0.3528367877006531	0.12877780199050903	0.18764065206050873	0.4161965548992157	0.0001
33	0.35152381658554077	0.12729588150978088	0.18640562891960144	0.4225224256515503	0.0001
34	0.35195404291152954	0.12629321217536926	0.18677598237991333	0.4202421009540558	0.0001
35	0.35178276896476746	0.12782610952854156	0.18657900393009186	0.42147499322891235	0.0001
36	0.35231974720954895	0.12849368155002594	0.18713095784187317	0.4192589223384857	0.0001
37	0.3515876829624176	0.12726719677448273	0.18659605085849762	0.4216739237308502	0.0001
38	0.35274896025657654	0.12742024660110474	0.1878250390291214	0.41570571064949036	0.0001
39	0.35124146938323975	0.12662582099437714	0.18624022603034973	0.4241558611392975	0.0001
40	0.35209622979164124	0.13019172847270966	0.18663105368614197	0.4224165081977844	0.0001
41	0.35067644715309143	0.1265629082918167	0.18584123253822327	0.42641735076904297	0.0001
42	0.3512935936450958	0.12775851786136627	0.18596960604190826	0.4262687563896179	0.0001
43	0.3510710895061493	0.12741515040397644	0.1859511435031891	0.42624664306640625	0.0001
44	0.35139599442481995	0.12441141903400421	0.18587811291217804	0.4265681505203247	0.0001
45	0.35247302055358887	0.127328023314476	0.18626871705055237	0.42492759227752686	0.0001
46	0.3505423069000244	0.12581512331962585	0.18559609353542328	0.4274958372116089	0.0001
47	0.3517468571662903	0.1250177025794983	0.18658187985420227	0.4231443405151367	0.0001
48	0.35043978691101074	0.12591718137264252	0.18564504384994507	0.42857199907302856	0.0001
49	0.35074281692504883	0.12717720866203308	0.1857146918773651	0.4274061322212219	0.0001
50	0.3515849709510803	0.1283276230096817	0.1857057511806488	0.42797213792800903	0.0001
51	0.35289809107780457	0.1288221776485443	0.1866857409477234	0.42265036702156067	0.0001
52	0.3505743443965912	0.12677451968193054	0.18569740653038025	0.4281761944293976	0.0001
53	0.35052910447120667	0.1273086667060852	0.18561594188213348	0.4285990595817566	0.0001
54	0.35016006231307983	0.12655657529830933	0.1853920817375183	0.4299810826778412	0.0001
55	0.35006165504455566	0.12509843707084656	0.18541744351387024	0.42986157536506653	0.0001
56	0.35072585940361023	0.12430255115032196	0.18582786619663239	0.4273306131362915	0.0001
57	0.3508891463279724	0.12534378468990326	0.18598994612693787	0.4273567497730255	0.0001
58	0.3492669463157654	0.12510134279727936	0.1846422404050827	0.4338167607784271	0.0001
59	0.3500733971595764	0.12414979934692383	0.18548892438411713	0.42818644642829895	0.0001
60	0.350059449672699	0.12591439485549927	0.18521927297115326	0.43032628297805786	0.0001
61	0.34978389739990234	0.126389279961586	0.18503333628177643	0.43050628900527954	0.0001
62	0.34984564781188965	0.1265084147453308	0.18499605357646942	0.4322562515735626	0.0001
63	0.35018646717071533	0.1270289421081543	0.18513011932373047	0.4321424067020416	0.0001
64	0.34996479749679565	0.12558279931545258	0.1853456199169159	0.43004974722862244	0.0001
65	0.3501463234424591	0.1280103474855423	0.1853969395160675	0.42989540100097656	1e-05
66	0.34930846095085144	0.1253172904253006	0.1847212016582489	0.43362313508987427	1e-05
67	0.3493542969226837	0.12613731622695923	0.18472003936767578	0.43344247341156006	1e-05
68	0.3500206172466278	0.12607118487358093	0.18558326363563538	0.4291488826274872	1e-05

Framework Versions

Transformers: 4.41.0
Pytorch: 2.5.0+cu124
Datasets: 3.0.2
Tokenizers: 0.19.1

lombardata
/

drone-DinoVdeau-produttoria_binary-probabilities-large-2024_11_03-batch-size64_freeze_probs