<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
finetuned-ViT-Indian-Food-Classification-v1
This model is a fine-tuned version of google/vit-base-patch16-224-in21k on the Human_Action_Recognition dataset. It achieves the following results on the evaluation set:
- Loss: 0.2665
- Accuracy: 0.9341
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0002
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 1.2019 | 0.3 | 100 | 0.9317 | 0.8555 |
| 0.6664 | 0.6 | 200 | 0.5432 | 0.8959 |
| 0.5096 | 0.9 | 300 | 0.4700 | 0.8990 |
| 0.6116 | 1.2 | 400 | 0.4504 | 0.8799 |
| 0.4326 | 1.5 | 500 | 0.3856 | 0.8980 |
| 0.3349 | 1.8 | 600 | 0.3471 | 0.9129 |
| 0.5141 | 2.1 | 700 | 0.3708 | 0.9033 |
| 0.32 | 2.4 | 800 | 0.3338 | 0.9139 |
| 0.2611 | 2.7 | 900 | 0.3159 | 0.9171 |
| 0.1836 | 3.0 | 1000 | 0.2696 | 0.9299 |
| 0.2492 | 3.3 | 1100 | 0.2979 | 0.9214 |
| 0.1846 | 3.6 | 1200 | 0.3165 | 0.9203 |
| 0.1505 | 3.9 | 1300 | 0.2806 | 0.9288 |
| 0.1854 | 4.2 | 1400 | 0.2665 | 0.9341 |
| 0.124 | 4.5 | 1500 | 0.2695 | 0.9341 |
| 0.0719 | 4.8 | 1600 | 0.2668 | 0.9320 |
Framework versions
- Transformers 4.21.2
- Pytorch 1.12.1+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1
