GLiT: Neural Architecture Search for Global and Local Image Transformer

Code for GLiT: Neural Architecture Search for Global and Local Image Transformer accepted by ICCV2021

Requirements

torch==1.7.0
torchvision==0.8.1
timm==0.3.2

The requirements.txt file lists other Python libraries that this project depends on, and they will be installed using: pip3 install -r requirements.txt

Results

method	FLOPs	Params	Top-1	model
GLiT-Tiny	1.33G	7.3M	76.4	Glit_Tiny

Training

python -m torch.distributed.launch --nproc_per_node=4 --use_env main.py --model glit_tiny_patch16_224 --clip-grad 1.0 --batch-size 256 --data-path /path/to/imagenet --output_dir /path/to/save

Evaluation

Download the model Glit_Tiny

python main.py --eval --resume glit_tiny.pth.tar --data-path /path/to/imagenet

Thanks

This implementation is based on DeiT. Please ref to their reposity for more details.

Citation

If you find that this project helps your research, please consider citing our paper:

@inproceedings{chen2021glit,
  title={Glit: Neural architecture search for global and local image transformer},
  author={Chen, Boyu and Li, Peixia and Li, Chuming and Li, Baopu and Bai, Lei and Lin, Chen and Sun, Ming and Yan, Junjie and Ouyang, Wanli},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={12--21},
  year={2021}
}