finetuned-vit-doc-text-classifer
This model is a fine-tuned version of google/vit-base-patch16-224-in21k on the ernie-ai/image-text-examples-ar-cn-latin-notext dataset. It achieves the following results on the evaluation set:
- Loss: 0.3107
- Accuracy: 0.9030
Model description
It is an image classificatin model fine-tuned to predict whether an images contains text and if that text is Latin script, Chinese or Arabic. It also classifies non-text images.
Training and evaluation data
Dataset: [ernie-ai/image-text-examples-ar-cn-latin-notext]
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: 8
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 0.2719 | 2.08 | 100 | 0.4120 | 0.8657 |
| 0.1027 | 4.17 | 200 | 0.3907 | 0.8881 |
| 0.0723 | 6.25 | 300 | 0.3107 | 0.9030 |
Framework versions
- Transformers 4.26.0
- Pytorch 1.13.1+cu116
- Datasets 2.9.0
- Tokenizers 0.13.2
- Downloads last month
- 21
Evaluation results
- Accuracy on ernie-ai/image-text-examples-ar-cn-latin-notextself-reported0.903