Vision Tranformer for image classification

Image Classification with Vision Transformer for blog image selection

How to use

train

1. Set the folder location where the images are located for train (train.py: train_list [line. 30])
2. Rename image file or change method for get label (train.py: label_list [line. 31])
3. Run train.py

inference

1. Set the folder location where the weight is located (inference.py: weight_path [line. 15])
2. Set the folder location where the images are located for inference (inference.py: inference_list [line. 21])
3. Run inference.py

Requirements

• torch == 1.13.1+cu116
• torchaudio == 0.13.1+cu116
• torchvision == 0.14.1+cu116
• python == 3.6.9
• scikit-learn == 1.2.1
• Pillow == 9.4.0
• numpy == 1.24.2

Reference

• Vision Transformer and MLP-Mixer Architectures(Google[jax])
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• ViT-L_16.npz