This is a PyTorch implementation of the paper CycleMLP: A MLP-like Architecture for Dense Prediction.
- (29/01/2022) CycleMLP is accepted by ICLR 2022 as an oral presentation.:fire::fire::fire:
- (22/07/2021) Initial release.
We provide CycleMLP models pretrained on ImageNet 2012.
Model | Parameters | FLOPs | Top 1 Acc. | Download |
---|---|---|---|---|
CycleMLP-B1 | 15M | 2.1G | 78.9% | model |
CycleMLP-B2 | 27M | 3.9G | 81.6% | model |
CycleMLP-B3 | 38M | 6.9G | 82.4% | model |
CycleMLP-B4 | 52M | 10.1G | 83.0% | model |
CycleMLP-B5 | 76M | 12.3G | 83.2% | model |
CycleMLP-T | 28M | 4.4G | 81.3% | model |
CycleMLP-S | 50M | 8.5G | 82.9% | model |
CycleMLP-B | 88M | 15.2G | 83.4% | model |
- PyTorch 1.7.0 and torchvision 0.8.1
- timm:
pip install 'git https://github.com/rwightman/pytorch-image-models@c2ba229d995c33aaaf20e00a5686b4dc857044be'
or
git clone https://github.com/rwightman/pytorch-image-models
cd pytorch-image-models
git checkout c2ba229d995c33aaaf20e00a5686b4dc857044be
pip install -e .
- fvcore (optional, for FLOPs calculation)
- mmcv, mmdetection, mmsegmentation (optional)
Download and extract ImageNet train and val images from http://image-net.org/. The directory structure is:
│path/to/imagenet/
├──train/
│ ├── n01440764
│ │ ├── n01440764_10026.JPEG
│ │ ├── n01440764_10027.JPEG
│ │ ├── ......
│ ├── ......
├──val/
│ ├── n01440764
│ │ ├── ILSVRC2012_val_00000293.JPEG
│ │ ├── ILSVRC2012_val_00002138.JPEG
│ │ ├── ......
│ ├── ......
To evaluate a pre-trained CycleMLP-B5 on ImageNet val with a single GPU run:
python main.py --eval --model CycleMLP_B5 --resume path/to/CycleMLP_B5.pth --data-path /path/to/imagenet
To train CycleMLP-B5 on ImageNet on a single node with 8 gpus for 300 epochs run:
python -m torch.distributed.launch --nproc_per_node=8 --use_env main.py --model CycleMLP_B5 --batch-size 128 --data-path /path/to/imagenet --output_dir /path/to/save
This code is based on DeiT and pytorch-image-models. Thanks for their wonderful works
@inproceedings{
chen2022cyclemlp,
title={Cycle{MLP}: A {MLP}-like Architecture for Dense Prediction},
author={Shoufa Chen and Enze Xie and Chongjian GE and Runjian Chen and Ding Liang and Ping Luo},
booktitle={International Conference on Learning Representations},
year={2022},
url={https://openreview.net/forum?id=NMEceG4v69Y}
}
CycleMLP is released under MIT License.