|
|
|
--- |
|
language: |
|
- eng |
|
license: wtfpl |
|
tags: |
|
- multilabel-image-classification |
|
- multilabel |
|
- generated_from_trainer |
|
base_model: facebook/dinov2-large |
|
model-index: |
|
- name: drone-DinoVdeau-large-2024_07_31-batch-size8_epochs100_freeze |
|
results: [] |
|
--- |
|
|
|
DinoVd'eau is a fine-tuned version of [facebook/dinov2-large](https://huggingface.co/facebook/dinov2-large). It achieves the following results on the test set: |
|
|
|
- Explained variance: 0.3552 |
|
- Loss: 0.3286 |
|
- MAE: 0.1261 |
|
- MSE: 0.0374 |
|
- R2: 0.3545 |
|
- RMSE: 0.1933 |
|
|
|
--- |
|
|
|
# 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](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 number of images for each class are given in the following table: |
|
| Class | train | val | test | Total | |
|
|:------------------------|--------:|------:|-------:|--------:| |
|
| Acropore_branched | 2371 | 782 | 785 | 3938 | |
|
| Acropore_digitised | 1693 | 580 | 579 | 2852 | |
|
| Acropore_sub_massive | 353 | 99 | 97 | 549 | |
|
| Acropore_tabular | 1112 | 420 | 410 | 1942 | |
|
| Algae | 13150 | 4386 | 4405 | 21941 | |
|
| Dead_coral | 6824 | 2242 | 2250 | 11316 | |
|
| Millepore | 1543 | 611 | 631 | 2785 | |
|
| No_acropore_encrusting | 2799 | 1044 | 1041 | 4884 | |
|
| No_acropore_massive | 6578 | 2216 | 2170 | 10964 | |
|
| No_acropore_sub_massive | 5252 | 1802 | 1793 | 8847 | |
|
| Rock | 13532 | 4529 | 4529 | 22590 | |
|
| Rubble | 12641 | 4222 | 4231 | 21094 | |
|
| Sand | 13315 | 4438 | 4438 | 22191 | |
|
|
|
--- |
|
|
|
# Training procedure |
|
|
|
## Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
|
|
- **Number of Epochs**: 100 |
|
- **Learning Rate**: 0.001 |
|
- **Train Batch Size**: 8 |
|
- **Eval Batch Size**: 8 |
|
- **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 | Explained Variance | Validation Loss | MAE | MSE | R2 | RMSE | Learning Rate |
|
--- | --- | --- | --- | --- | --- | --- | --- |
|
0 | 0 | 0 | 0 | 0 | 0 | 0 | 0.001 |
|
1 | 0.233 | 0.353 | 0.135 | 0.045 | 0.228 | 0.212 | 0.001 |
|
2 | 0.242 | 0.354 | 0.145 | 0.046 | 0.232 | 0.213 | 0.001 |
|
3 | 0.264 | 0.35 | 0.141 | 0.044 | 0.26 | 0.209 | 0.001 |
|
4 | 0.273 | 0.347 | 0.14 | 0.043 | 0.271 | 0.208 | 0.001 |
|
5 | 0.257 | 0.365 | 0.139 | 0.043 | 0.256 | 0.208 | 0.001 |
|
6 | 0.228 | 0.37 | 0.14 | 0.044 | 0.226 | 0.209 | 0.001 |
|
7 | 0.28 | 0.357 | 0.137 | 0.042 | 0.278 | 0.206 | 0.001 |
|
8 | 0.287 | 0.345 | 0.134 | 0.042 | 0.282 | 0.205 | 0.001 |
|
9 | 0.28 | 0.345 | 0.139 | 0.043 | 0.278 | 0.206 | 0.001 |
|
10 | 0.282 | 0.386 | 0.133 | 0.043 | 0.279 | 0.206 | 0.001 |
|
11 | 0.28 | 0.347 | 0.137 | 0.043 | 0.276 | 0.207 | 0.001 |
|
12 | 0.285 | 0.351 | 0.137 | 0.042 | 0.28 | 0.206 | 0.001 |
|
13 | 0.277 | 0.397 | 0.136 | 0.043 | 0.275 | 0.207 | 0.001 |
|
14 | 0.276 | 0.359 | 0.137 | 0.043 | 0.271 | 0.208 | 0.001 |
|
15 | 0.276 | 0.349 | 0.136 | 0.042 | 0.274 | 0.206 | 0.001 |
|
16 | 0.303 | 0.339 | 0.133 | 0.041 | 0.303 | 0.202 | 0.0001 |
|
17 | 0.306 | 0.338 | 0.134 | 0.04 | 0.306 | 0.201 | 0.0001 |
|
18 | 0.31 | 0.337 | 0.132 | 0.04 | 0.308 | 0.2 | 0.0001 |
|
19 | 0.307 | 0.338 | 0.131 | 0.04 | 0.305 | 0.201 | 0.0001 |
|
20 | 0.312 | 0.336 | 0.131 | 0.04 | 0.311 | 0.2 | 0.0001 |
|
21 | 0.312 | 0.337 | 0.129 | 0.04 | 0.308 | 0.2 | 0.0001 |
|
22 | 0.316 | 0.336 | 0.132 | 0.04 | 0.315 | 0.199 | 0.0001 |
|
23 | 0.32 | 0.335 | 0.131 | 0.039 | 0.319 | 0.199 | 0.0001 |
|
24 | 0.319 | 0.335 | 0.13 | 0.04 | 0.318 | 0.199 | 0.0001 |
|
25 | 0.324 | 0.334 | 0.13 | 0.039 | 0.323 | 0.198 | 0.0001 |
|
26 | 0.32 | 0.335 | 0.13 | 0.04 | 0.318 | 0.199 | 0.0001 |
|
27 | 0.321 | 0.335 | 0.13 | 0.039 | 0.32 | 0.198 | 0.0001 |
|
28 | 0.326 | 0.334 | 0.127 | 0.039 | 0.321 | 0.198 | 0.0001 |
|
29 | 0.33 | 0.333 | 0.129 | 0.039 | 0.328 | 0.197 | 0.0001 |
|
30 | 0.33 | 0.333 | 0.13 | 0.039 | 0.33 | 0.197 | 0.0001 |
|
31 | 0.328 | 0.333 | 0.13 | 0.039 | 0.325 | 0.198 | 0.0001 |
|
32 | 0.331 | 0.332 | 0.128 | 0.039 | 0.33 | 0.197 | 0.0001 |
|
33 | 0.334 | 0.333 | 0.13 | 0.039 | 0.331 | 0.196 | 0.0001 |
|
34 | 0.33 | 0.333 | 0.129 | 0.039 | 0.328 | 0.197 | 0.0001 |
|
35 | 0.325 | 0.334 | 0.131 | 0.039 | 0.324 | 0.198 | 0.0001 |
|
36 | 0.337 | 0.332 | 0.13 | 0.038 | 0.337 | 0.196 | 0.0001 |
|
37 | 0.328 | 0.334 | 0.13 | 0.039 | 0.327 | 0.197 | 0.0001 |
|
38 | 0.338 | 0.332 | 0.129 | 0.038 | 0.336 | 0.196 | 0.0001 |
|
39 | 0.338 | 0.332 | 0.128 | 0.038 | 0.338 | 0.196 | 0.0001 |
|
40 | 0.337 | 0.332 | 0.129 | 0.038 | 0.336 | 0.196 | 0.0001 |
|
41 | 0.335 | 0.333 | 0.131 | 0.039 | 0.333 | 0.196 | 0.0001 |
|
42 | 0.338 | 0.332 | 0.129 | 0.038 | 0.337 | 0.196 | 0.0001 |
|
43 | 0.338 | 0.331 | 0.129 | 0.038 | 0.338 | 0.196 | 0.0001 |
|
44 | 0.336 | 0.333 | 0.128 | 0.039 | 0.335 | 0.196 | 0.0001 |
|
45 | 0.339 | 0.331 | 0.128 | 0.038 | 0.338 | 0.196 | 0.0001 |
|
46 | 0.341 | 0.332 | 0.129 | 0.038 | 0.339 | 0.195 | 0.0001 |
|
47 | 0.34 | 0.331 | 0.127 | 0.038 | 0.339 | 0.196 | 0.0001 |
|
48 | 0.299 | 0.339 | 0.131 | 0.039 | 0.295 | 0.199 | 0.0001 |
|
49 | 0.338 | 0.331 | 0.128 | 0.038 | 0.337 | 0.196 | 0.0001 |
|
50 | 0.342 | 0.332 | 0.127 | 0.038 | 0.339 | 0.196 | 0.0001 |
|
51 | 0.341 | 0.331 | 0.127 | 0.038 | 0.341 | 0.195 | 0.0001 |
|
52 | 0.345 | 0.33 | 0.127 | 0.038 | 0.344 | 0.195 | 0.0001 |
|
53 | 0.34 | 0.331 | 0.128 | 0.038 | 0.339 | 0.196 | 0.0001 |
|
54 | 0.341 | 0.331 | 0.129 | 0.038 | 0.34 | 0.196 | 0.0001 |
|
55 | 0.349 | 0.329 | 0.127 | 0.038 | 0.349 | 0.194 | 0.0001 |
|
56 | 0.344 | 0.33 | 0.126 | 0.038 | 0.343 | 0.195 | 0.0001 |
|
57 | 0.341 | 0.331 | 0.126 | 0.038 | 0.339 | 0.196 | 0.0001 |
|
58 | 0.348 | 0.33 | 0.126 | 0.038 | 0.347 | 0.194 | 0.0001 |
|
59 | 0.343 | 0.332 | 0.128 | 0.038 | 0.341 | 0.195 | 0.0001 |
|
60 | 0.346 | 0.331 | 0.128 | 0.038 | 0.345 | 0.195 | 0.0001 |
|
61 | 0.346 | 0.33 | 0.125 | 0.038 | 0.344 | 0.195 | 0.0001 |
|
62 | 0.347 | 0.329 | 0.126 | 0.038 | 0.346 | 0.194 | 1e-05 |
|
63 | 0.35 | 0.33 | 0.128 | 0.038 | 0.348 | 0.194 | 1e-05 |
|
64 | 0.345 | 0.33 | 0.126 | 0.038 | 0.344 | 0.195 | 1e-05 |
|
65 | 0.349 | 0.33 | 0.128 | 0.038 | 0.347 | 0.195 | 1e-05 |
|
|
|
|
|
--- |
|
|
|
# CO2 Emissions |
|
|
|
The estimated CO2 emissions for training this model are documented below: |
|
|
|
- **Emissions**: 0.19095786836275294 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 |
|
|
|
|