jittor-image-models / fairscale

PyTorch extensions for high performance and large scale training.

Jittor Image Models

Jittor Image Models (jimm) is a library for pulling together a wide variety of SOTA deep learning models in the Jittor framework. Based on jimm, we achieved the first place of the Dog Species Classification track in the Jittor AI Competition in 2021.

Our jimm is modified from PyTorch Image Models (timm) which helps fine-tune PyTorch models list systematically by timm in Jittor.

More specifically, PyTorch Image Models (timm) is an excellent project created by Ross Wightman and perfected by many outstanding contributors. Details about timm is available at: https://github.com/rwightman/pytorch-image-models

Jittor is a high-performance deep learning framework based on JIT compiling and meta-operators. More details about Jittor can be found via: https://github.com/Jittor/jittor

In our jimm, we reproduce part of pytorch-image-models in the Jittor deep learning framework, and also provide a training demo to make it easier for you to get started.

Update News

Feb 25, 2022

• Add VAN, VAN pretrained models can be download from https://github.com/Visual-Attention-Network/VAN-Classification.
• You have to transfer the download .pth file by following:
  model = torch.load('van_base_828.pth', map_location=torch.device('cpu'))
  torch.save(model['state_dict'],'van_base.pth')

Sep 06, 2021

• Add VOLO, Swin Transformer, EfficientNet-V2
• VOLO pretrained models can be download from https://github.com/sail-sg/volo.

June 10, 2021

• Add HRNet models.

Currently Supported Models

• DeiT (Vision Transformer) - https://arxiv.org/abs/2012.12877
• EfficientNet
  • EfficientNet (B0-B7) - https://arxiv.org/abs/1905.11946
  • EfficientNet AdvProp (B0-B8) - https://arxiv.org/abs/1911.09665
  • EfficientNet NoisyStudent (B0-B7, L2) - https://arxiv.org/abs/1911.04252
  • EfficientNet V2 - https://arxiv.org/abs/2104.00298
• HRNet - https://arxiv.org/abs/1908.07919
• ResNet/ResNeXt
  • ResNet (v1b/v1.5) - https://arxiv.org/abs/1512.03385
  • ResNeXt - https://arxiv.org/abs/1611.05431
  • Weakly-supervised (WSL) Instagram pretrained / ImageNet tuned ResNeXt101 - https://arxiv.org/abs/1805.00932
  • Semi-supervised (SSL) / Semi-weakly Supervised (SWSL) ResNet/ResNeXts - https://arxiv.org/abs/1905.00546
• Swin Transformer - https://arxiv.org/abs/2103.14030
• VAN - https://arxiv.org/abs/2202.09741
• ViT - https://arxiv.org/abs/2010.11929
• VOLO - https://arxiv.org/abs/2106.13112

More models provided by timm will continue to be updated.

Results

Model validation results can be found in the following url: https://rwightman.github.io/pytorch-image-models/results/

TODO List

• RepVGG - https://arxiv.org/abs/2101.03697
• Big Transfer ResNetV2 (BiT) - https://arxiv.org/abs/1912.11370
• NFNet-F - https://arxiv.org/abs/2102.06171

Contacts

If you have any questions about our work, please do not hesitate to contact us by emails.

Xuhao Sun: [email protected]

Yang Shen: [email protected]

Xiu-Shen Wei (Primary contact): [email protected]