RepVGG-openMMLab

This repository is the code that needs to be submitted for OpenMMLab Algorithm Ecological Challenge，the paper is RepVGG: Making VGG-style ConvNets Great Again

Result :)

1. RepVGG-b2g4 top1 is 79.78
2. RepVGG-b3g4 top1 is 80.31
3. RepVGG-B3 top1 is 80.59

How TO Use?

What is MMCV?

MMCV is a foundational library for computer vision research and supports many research projects as below:

• MMClassification:OpenMMLab image classification toolbox and benchmark.
• MMDetection:OpenMMLab detection toolbox and benchmark.
• MMDetection3D:OpenMMLab’s next-generation platform for general 3D object detection.
• MMSegmentation:OpenMMLab semantic segmentation toolbox and benchmark.
• MMAction2:OpenMMLab’s next-generation action understanding toolbox and benchmark.
• MMTracking:OpenMMLab video perception toolbox and benchmark.
• ...

Using MMCV for the first time??

1. Install MMCV using MIM

pip install git+https://github.com/open-mmlab/mim.git

mim install mmcv-full

2. clone MMClassification and install

git clone https://github.com/open-mmlab/mmclassification.git

cd mmclassification

pip install -e .

3. Register RepVGG in MMclassification

cp RepVGG-openMMLab/backbones/RepVGG.py \
mmclassification/mmcls/models/backbones/

in mmclassification/mmcls/models/backbones/__init__.py

...
from .RepVGG import RepVGG

__all__ = [
    ..., 'RepVGG'
]

4. copy config file to mmclassification/config

cp RepVGG-openMMLab/config/repvggb2g4_b32x8.py \
mmclassification/config/

5. Train Model(If you downloaded Imagenet)

cd mmclassification
python tools/train.py config/repvggb2g4_b32x8.py 

Download && Unzip ImageNet

data/download_imagenet.sh ，This script can automatically build the file structure that Imagenet needs for mmcls

mkdir -p mmclassification/data
cp RepVGG-openMMLab/data/download_imagenet.sh mmclassification/data
cd mmclassification/data
bash download_imagenet.sh

Use the pre-trained model on Google Drive

pre-trained model

• RepVGGB2g4.pth

• RepVGGB3g4.pth

• RepVGGB3.pth

Test Model

in mmclassification

python tools/test.py config/repvggb2g4_b32x8.py ${repvggb2g4 model file} --meterics accuracy

someting like this..

reference

1. RepVGG：极简架构，SOTA性能，让VGG式模型再次伟大（CVPR-2021)
2. RepVGG: Making VGG-style ConvNets Great Again (CVPR-2021) (PyTorch)
3. MMClassification Docs
4. ImageNet
5. MMCV DOCs