This repository is the official implementation of "RMNet: Equivalently Removing Residual Connection from Networks".
To install requirements:
pip install torch
pip install torchvision
To train the models in the paper, run this command:
python train.py -a rmrep_69 --dist-url 'tcp://127.0.0.1:23333' --dist-backend 'nccl' --multiprocessing-distributed --world-size 1 --rank 0 --workers 32 [imagenet-folder with train and val folders]
You can download pretrained models here:
- Download our pre-trained models trained on ImageNet from Google Drive or Baidu Cloud(提取码:0mto).
To evaluate our pre-trained models trained on ImageNet, run:
python train.py -a rmrep_69 -e checkpoint/rmrep_69.pth.tar [imagenet-folder with train and val folders]
Our model achieves the following performance on :
| Arch | Top-1 Accuracy(%) | Top-5 Accuracy(%) | Train FLOPs(G) | Test FLOPs(M) |
|---|---|---|---|---|
| RepVGG-21 | 72.508 | 90.840 | 2.4 | 2.1 |
| RepVGG-21(RM 0.25) | 72.590 | 90.924 | 2.1 | 2.1 |
| RepVGG-37 | 74.408 | 91.900 | 4.4 | 4.0 |
| RepVGG-37(RM 0.25) | 74.478 | 91.892 | 3.9 | 4.0 |
| RepVGG-69 | 74.526 | 92.182 | 8.6 | 7.7 |
| RepVGG-69(RM 0.5) | 75.088 | 92.144 | 6.5 | 7.7 |
| RepVGG-133 | 70.912 | 89.788 | 16.8 | 15.1 |
| RepVGG-133(RM 0.75) | 74.560 | 92.000 | 10.6 | 15.1 |
| Model name | Top 1 Accuracy(%) | Top 5 Accuracy(%) |
|---|---|---|
| RMNeXt 41x5_16 | 78.498 | 94.086 |
| RMNeXt 50x5_32 | 79.076 | 94.444 |
| RMNeXt 50x6_32 | 79.57 | 94.644 |
| RMNeXt 101x6_16 | 80.07 | 94.918 |
| RMNeXt 152x6_32 | 80.356 | 80.356 |
| Arch | Top-1 Accuracy(%) | Speed() |
|---|---|---|
| RepVGG A0 | 72.41 | 3256 |
| RMNet A0 | TODO | 3256 |
| RepVGG A1 | 74.46 | 2339 |
| RMNet A1 | TODO | 2339 |
| RepVGG A2 | 76.48 | 1322 |
| RMNet A2 | TODO | 1322 |
| RepVGG B0 | 75.14 | 1817 |
| RMNet B0 | TODO | 1817 |
| RepVGG B1 | 78.37 | 685 |
| RMNet B1 | TODO | 685 |
| RepVGG B2 | 78.78 | 460 |
| RMNet B2 | TODO | 460 |
If you find this code useful, please cite the following paper:
@misc{meng2021rmnet,
title={RMNet: Equivalently Removing Residual Connection from Networks},
author={Fanxu Meng and Hao Cheng and Jiaxin Zhuang and Ke Li and Xing Sun},
year={2021},
eprint={2111.00687},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
Our code is based on RepVGG