Softmax-free Linear Transformers,
Jiachen Lu, Li Zhang, Junge Zhang, Xiatian Zhu, Hang Xu, Jianfeng Feng

1. We propose a normalized softmax-free self-attention with stronger generalizability.
2. SOFT is now avaliable on more vision tasks (object detection and semantic segmentation).

• timm==0.3.2

• torch>=1.7.0 and torchvision that matches the PyTorch installation

• cuda>=10.2

Compilation may be fail on cuda < 10.2.
We have compiled it successfully on cuda 10.2 and cuda 11.2.

Download and extract ImageNet train and val images from http://image-net.org/. The directory structure is the standard layout for the torchvision datasets.ImageFolder, and the training and validation data is expected to be in the train/ folder and val folder respectively:

/path/to/imagenet/
  train/
    class1/
      img1.jpeg
    class2/
      img2.jpeg
  val/
    class1/
      img3.jpeg
    class/2
      img4.jpeg

git clone https://github.com/fudan-zvg/SOFT.git
python -m pip install -e SOFT

We have two implementations of Gaussian Kernel: PyTorch version and the exact form of Gaussian function implemented by cuda. The config file containing cuda is the cuda implementation. Both implementations yield same performance. Please install SOFT before running the cuda version.

./dist_train.sh ${GPU_NUM} --data ${DATA_PATH} --config ${CONFIG_FILE}
# For example, train SOFT-Tiny on Imagenet training dataset with 8 GPUs
./dist_train.sh 8 --data ${DATA_PATH} --config config/SOFT_Tiny.yaml

./dist_train.sh ${GPU_NUM} --data ${DATA_PATH} --config ${CONFIG_FILE} --eval_checkpoint ${CHECKPOINT_FILE} --eval

# For example, test SOFT-Tiny on Imagenet validation dataset with 8 GPUs

./dist_train.sh 8 --data ${DATA_PATH} --config config/SOFT_Tiny.yaml --eval_checkpoint ${CHECKPOINT_FILE} --eval

@inproceedings{SOFT,
    title={SOFT: Softmax-free Transformer with Linear Complexity}, 
    author={Lu, Jiachen and Yao, Jinghan and Zhang, Junge and Zhu, Xiatian and Xu, Hang and Gao, Weiguo and Xu, Chunjing and Xiang, Tao and Zhang, Li},
    booktitle={NeurIPS},
    year={2021}
}

MIT

Thanks to previous open-sourced repo:
Detectron2
T2T-ViT
PVT
Nystromformer
pytorch-image-models