--- language: - eng license: cc0-1.0 tags: - multilabel-image-classification - multilabel - generated_from_trainer base_model: drone-DinoVdeau-produttoria-probabilities-large-2024_11_04-batch-size64_freeze_probs model-index: - name: drone-DinoVdeau-produttoria-probabilities-large-2024_11_04-batch-size64_freeze_probs results: [] --- drone-DinoVdeau-produttoria-probabilities is a fine-tuned version of [drone-DinoVdeau-produttoria-probabilities-large-2024_11_04-batch-size64_freeze_probs](https://huggingface.co/drone-DinoVdeau-produttoria-probabilities-large-2024_11_04-batch-size64_freeze_probs). It achieves the following results on the test set: - Loss: 0.3194 - F1 Micro: 0.8663 - F1 Macro: 0.8311 - Accuracy: 0.1799 - RMSE: 0.2404 - MAE: 0.1536 - R2: 0.4282 | Class | F1 per class | |----------|-------| | Acropore_branched | 0.8010 | | Acropore_digitised | 0.7454 | | Acropore_tabular | 0.6426 | | Algae | 0.9852 | | Dead_coral | 0.8448 | | Fish | 0.7497 | | Millepore | 0.6641 | | No_acropore_encrusting | 0.7391 | | No_acropore_massive | 0.8688 | | No_acropore_sub_massive | 0.8137 | | Rock | 0.9924 | | Rubble | 0.9691 | | Sand | 0.9888 | --- # Model description drone-DinoVdeau-produttoria-probabilities is a model built on top of drone-DinoVdeau-produttoria-probabilities-large-2024_11_04-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](https://github.com/SeatizenDOI/DinoVdeau). - **Developed by:** [lombardata](https://huggingface.co/lombardata), credits to [César Leblanc](https://huggingface.co/CesarLeblanc) and [Victor Illien](https://huggingface.co/groderg) --- # 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 | 2028 | 684 | 686 | 3398 | | Acropore_digitised | 2006 | 735 | 717 | 3458 | | Acropore_tabular | 1237 | 461 | 451 | 2149 | | Algae | 11086 | 3671 | 3675 | 18432 | | Dead_coral | 6354 | 2161 | 2147 | 10662 | | Fish | 4032 | 1430 | 1430 | 6892 | | Millepore | 1943 | 783 | 772 | 3498 | | No_acropore_encrusting | 2663 | 986 | 957 | 4606 | | No_acropore_massive | 6897 | 2375 | 2375 | 11647 | | No_acropore_sub_massive | 5416 | 1988 | 1958 | 9362 | | Rock | 11164 | 3726 | 3725 | 18615 | | Rubble | 10687 | 3570 | 3572 | 17829 | | Sand | 11151 | 3726 | 3723 | 18600 | --- # Training procedure ## Training hyperparameters The following hyperparameters were used during training: - **Number of Epochs**: 69.0 - **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 | 0.36101797223091125 | 0.1878 | 0.2645 | 0.2818 | 0.001 2 | 0.3464529514312744 | 0.1778 | 0.2566 | 0.3349 | 0.001 3 | 0.34133487939834595 | 0.1731 | 0.2532 | 0.3536 | 0.001 4 | 0.3406243324279785 | 0.1743 | 0.2532 | 0.3519 | 0.001 5 | 0.3341675400733948 | 0.1661 | 0.2496 | 0.3702 | 0.001 6 | 0.33848145604133606 | 0.1739 | 0.2512 | 0.3651 | 0.001 7 | 0.3320676386356354 | 0.1650 | 0.2476 | 0.3836 | 0.001 8 | 0.3332081437110901 | 0.1629 | 0.2484 | 0.3802 | 0.001 9 | 0.3305376172065735 | 0.1652 | 0.2468 | 0.3859 | 0.001 10 | 0.33136793971061707 | 0.1655 | 0.2476 | 0.3827 | 0.001 11 | 0.3319685757160187 | 0.1602 | 0.2474 | 0.3840 | 0.001 12 | 0.3341500759124756 | 0.1683 | 0.2494 | 0.3761 | 0.001 13 | 0.33248215913772583 | 0.1649 | 0.2480 | 0.3821 | 0.001 14 | 0.33228376507759094 | 0.1700 | 0.2472 | 0.3878 | 0.001 15 | 0.334873229265213 | 0.1703 | 0.2493 | 0.3749 | 0.001 16 | 0.3279329538345337 | 0.1649 | 0.2448 | 0.3983 | 0.0001 17 | 0.3279244005680084 | 0.1648 | 0.2448 | 0.3984 | 0.0001 18 | 0.3262367248535156 | 0.1622 | 0.2440 | 0.4025 | 0.0001 19 | 0.3247373402118683 | 0.1588 | 0.2432 | 0.4046 | 0.0001 20 | 0.32612329721450806 | 0.1625 | 0.2433 | 0.4059 | 0.0001 21 | 0.3241129517555237 | 0.1606 | 0.2424 | 0.4095 | 0.0001 22 | 0.32355180382728577 | 0.1587 | 0.2422 | 0.4111 | 0.0001 23 | 0.3242079019546509 | 0.1601 | 0.2423 | 0.4107 | 0.0001 24 | 0.3227241337299347 | 0.1586 | 0.2414 | 0.4150 | 0.0001 25 | 0.3223778307437897 | 0.1587 | 0.2413 | 0.4148 | 0.0001 26 | 0.3217927813529968 | 0.1557 | 0.2413 | 0.4143 | 0.0001 27 | 0.3227355182170868 | 0.1603 | 0.2416 | 0.4138 | 0.0001 28 | 0.32067713141441345 | 0.1562 | 0.2405 | 0.4186 | 0.0001 29 | 0.32205939292907715 | 0.1597 | 0.2411 | 0.4163 | 0.0001 30 | 0.32246074080467224 | 0.1608 | 0.2413 | 0.4164 | 0.0001 31 | 0.3223503530025482 | 0.1535 | 0.2416 | 0.4134 | 0.0001 32 | 0.3212696313858032 | 0.1553 | 0.2408 | 0.4180 | 0.0001 33 | 0.32156360149383545 | 0.1583 | 0.2414 | 0.4123 | 0.0001 34 | 0.3205103278160095 | 0.1562 | 0.2406 | 0.4172 | 0.0001 35 | 0.3197581171989441 | 0.1535 | 0.2399 | 0.4215 | 0.0001 36 | 0.3211075961589813 | 0.1577 | 0.2406 | 0.4187 | 0.0001 37 | 0.3203599154949188 | 0.1520 | 0.2403 | 0.4188 | 0.0001 38 | 0.32143038511276245 | 0.1560 | 0.2409 | 0.4170 | 0.0001 39 | 0.3195198178291321 | 0.1520 | 0.2397 | 0.4226 | 0.0001 40 | 0.3207896649837494 | 0.1577 | 0.2404 | 0.4204 | 0.0001 41 | 0.3197501003742218 | 0.1547 | 0.2398 | 0.4217 | 0.0001 42 | 0.32175716757774353 | 0.1589 | 0.2410 | 0.4174 | 0.0001 43 | 0.3189575970172882 | 0.1544 | 0.2396 | 0.4235 | 0.0001 44 | 0.31898385286331177 | 0.1534 | 0.2396 | 0.4230 | 0.0001 45 | 0.31977778673171997 | 0.1566 | 0.2397 | 0.4239 | 0.0001 46 | 0.3193351626396179 | 0.1556 | 0.2398 | 0.4213 | 0.0001 47 | 0.31895366311073303 | 0.1524 | 0.2393 | 0.4245 | 0.0001 48 | 0.3192996680736542 | 0.1525 | 0.2398 | 0.4215 | 0.0001 49 | 0.32073548436164856 | 0.1558 | 0.2405 | 0.4187 | 0.0001 50 | 0.3198453485965729 | 0.1572 | 0.2400 | 0.4218 | 1e-05 51 | 0.32436585426330566 | 0.1602 | 0.2426 | 0.4092 | 1e-05 52 | 0.31899821758270264 | 0.1550 | 0.2396 | 0.4227 | 1e-05 53 | 0.31892043352127075 | 0.1552 | 0.2394 | 0.4249 | 1e-05 54 | 0.3194037675857544 | 0.1540 | 0.2396 | 0.4227 | 1e-05 55 | 0.3184601366519928 | 0.1539 | 0.2391 | 0.4250 | 1e-05 56 | 0.318115234375 | 0.1527 | 0.2388 | 0.4273 | 1e-05 57 | 0.31871330738067627 | 0.1532 | 0.2392 | 0.4259 | 1e-05 58 | 0.32010164856910706 | 0.1567 | 0.2401 | 0.4217 | 1e-05 59 | 0.31807705760002136 | 0.1528 | 0.2388 | 0.4270 | 1e-05 60 | 0.3181913495063782 | 0.1534 | 0.2389 | 0.4256 | 1e-05 61 | 0.3185857832431793 | 0.1510 | 0.2391 | 0.4255 | 1e-05 62 | 0.32031872868537903 | 0.1596 | 0.2398 | 0.4240 | 1e-05 63 | 0.31964218616485596 | 0.1570 | 0.2397 | 0.4242 | 1e-05 64 | 0.31808170676231384 | 0.1527 | 0.2391 | 0.4244 | 1e-05 65 | 0.31850185990333557 | 0.1550 | 0.2390 | 0.4259 | 1e-05 66 | 0.3186076879501343 | 0.1562 | 0.2389 | 0.4278 | 1.0000000000000002e-06 67 | 0.3181016743183136 | 0.1526 | 0.2387 | 0.4270 | 1.0000000000000002e-06 68 | 0.3194774389266968 | 0.1549 | 0.2397 | 0.4221 | 1.0000000000000002e-06 69 | 0.3183264136314392 | 0.1540 | 0.2390 | 0.4259 | 1.0000000000000002e-06 --- # Framework Versions - **Transformers**: 4.41.0 - **Pytorch**: 2.5.0+cu124 - **Datasets**: 3.0.2 - **Tokenizers**: 0.19.1