DinoVd'eau is a fine-tuned version of microsoft/resnet-50. It achieves the following results on the test set:

  • Loss: nan
  • F1 Micro: 0.0002
  • F1 Macro: 0.0002
  • Roc Auc: 0.4995
  • Accuracy: 0.0003

Model description

DinoVd'eau is a model built on top of dinov2 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.


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

Class train val test Total
Acropore_branched 1469 464 475 2408
Acropore_digitised 568 160 160 888
Acropore_sub_massive 150 50 43 243
Acropore_tabular 999 297 293 1589
Algae_assembly 2546 847 845 4238
Algae_drawn_up 367 126 127 620
Algae_limestone 1652 557 563 2772
Algae_sodding 3148 984 985 5117
Atra/Leucospilota 1084 348 360 1792
Bleached_coral 219 71 70 360
Blurred 191 67 62 320
Dead_coral 1979 642 643 3264
Fish 2018 656 647 3321
Homo_sapiens 161 62 59 282
Human_object 157 58 55 270
Living_coral 406 154 141 701
Millepore 385 127 125 637
No_acropore_encrusting 441 130 154 725
No_acropore_foliaceous 204 36 46 286
No_acropore_massive 1031 336 338 1705
No_acropore_solitary 202 53 48 303
No_acropore_sub_massive 1401 433 422 2256
Rock 4489 1495 1473 7457
Rubble 3092 1030 1001 5123
Sand 5842 1939 1938 9719
Sea_cucumber 1408 439 447 2294
Sea_urchins 327 107 111 545
Sponge 269 96 105 470
Syringodium_isoetifolium 1212 392 391 1995
Thalassodendron_ciliatum 782 261 260 1303
Useless 579 193 193 965

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • Number of Epochs: 150
  • Learning Rate: 0.001
  • Train Batch Size: 32
  • Eval Batch Size: 32
  • 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 Accuracy F1 Macro F1 Micro Learning Rate
1 nan 0.0 0.0 0.0 0.001
2 nan 0.000693000693000693 0.00031409501374165687 0.00040576181781294376 0.001
3 nan 0.0017325017325017325 0.0007850525985241011 0.0010049241282283187 0.001
4 nan 0.0 0.0 0.0 0.001
5 nan 0.0010395010395010396 0.00047177229124076113 0.0006430178973314757 0.001
6 nan 0.0003465003465003465 0.00015712153350616704 0.000206782464846981 0.001
7 nan 0.0 0.0 0.0 0.0001
8 nan 0.0003465003465003465 0.00015710919088766695 0.0002061218179944347 0.0001
9 nan 0.0 0.0 0.0 0.0001
10 nan 0.000693000693000693 0.00031441597233139445 0.0004230565838180856 0.0001
11 nan 0.0 0.0 0.0 0.0001

CO2 Emissions

The estimated CO2 emissions for training this model are documented below:

  • Emissions: 0.12280230273705112 grams of CO2
  • Source: Code Carbon
  • Training Type: fine-tuning
  • Geographical Location: Brest, France
  • Hardware Used: NVIDIA Tesla V100 PCIe 32 Go

Framework Versions

  • Transformers: 4.41.1
  • Pytorch: 2.3.0+cu121
  • Datasets: 2.19.1
  • Tokenizers: 0.19.1
Downloads last month
36
Safetensors
Model size
23.6M params
Tensor type
F32
·
Inference API
Unable to determine this model's library. Check the docs .

Model tree for lombardata/resnet-50-2024_09_13-batch-size32_epochs150_freeze

Finetuned
(134)
this model