xuduo35 / imgcls_pytorch

image classification pytorch framework supporting SOTA backbones

imgcls_pytorch

A image classification pytorch framework supporting SOTA backbones

Prepare dataset

Put your images under a directory as below

• your_dataset_directory
  • class1
    • 1.jpg
    • 2.jpg
  • class2
    • 1.jpg
    • 2.jpg
    • ...
  • ...

Train

CUDA_VISIBLE_DEVICES=0 python3 -u train.py --backbone resnet101 --workers 32 --lr=0.001 --epochs 30 --train_bs 160 --datadir your_dataset_directory

Demo

CUDA_VISIBLE_DEVICES=0 PORT=8000 python3 -u app.py

Tensorboard displaying Chinese support

python3 fixfont.py

Follow its instructions.

Supported backbones

• alexnet
• resnet18,resnet34,resnet50,resnet101, resnet152, resnext101_32x4d, resnext101_64x4d
• vgg11_bn, vgg16_bn
• densenet121, densenet169, densenet161
• inceptionv3, inceptionv4, inceptionresnetv2, bninception
• xception, xception_att
• dpn98, dpn107, dpn131
• senet154, se_resnet50, se_resnet101, se_resnet152, se_resnext50_32x4d
• pnasnet5large
• polynet
• efficientnet, efficientnetV2