RepMLP: Re-parameterizing Convolutions into Fully-connected Layers for Image Recognition, arxiv

PaddlePaddle training/validation code and pretrained models for RepMLP.

The official pytorch implementation is here.

This implementation is developed by PaddleViT.

RepMLP Model Overview

Update

• Update (2021-09-27): Model FLOPs and # params are uploaded.
• Update (2021-09-14): Code is released and ported weights are uploaded.

Models Zoo

Model	Acc@1	Acc@5	#Params	FLOPs	Image Size	Crop_pct	Interpolation	Link
repmlp_res50_light_224	77.01	93.46	87.1M	3.3G	224	0.875	bicubic	google/baidu(b4fg)

*The results are evaluated on ImageNet2012 validation set.

Note: RepMLP weights are ported from here.

Notebooks

We provide a few notebooks in aistudio to help you get started:

*(coming soon)*

Requirements

• Python>=3.6
• yaml>=0.2.5
• PaddlePaddle>=2.1.0
• yacs>=0.1.8

Data

ImageNet2012 dataset is used in the following folder structure:

│imagenet/
├──train/
│  ├── n01440764
│  │   ├── n01440764_10026.JPEG
│  │   ├── n01440764_10027.JPEG
│  │   ├── ......
│  ├── ......
├──val/
│  ├── n01440764
│  │   ├── ILSVRC2012_val_00000293.JPEG
│  │   ├── ILSVRC2012_val_00002138.JPEG
│  │   ├── ......
│  ├── ......

Usage

To use the model with pretrained weights, download the .pdparam weight file and change related file paths in the following python scripts. The model config files are located in ./configs/.

For example, assume the downloaded weight file is stored in ./RepMLP-Res50-light-224_train.pdparams, to use the RepMLP-Res50-light-224_train model in python:

from config import get_config
from resmlp_resnet import build_resmlp_resnet as build_model
# config files in ./configs/
config = get_config('./configs/repmlpres50_light_224_train.yaml')
# build model
model = build_model(config)
# load pretrained weights, .pdparams is NOT needed
model_state_dict = paddle.load('./RepMLP-Res50-light-224_train')
model.set_dict(model_state_dict)

Evaluation

To evaluate ResMLP model performance on ImageNet2012 with a single GPU, run the following script using command line:

sh run_eval.sh

or

CUDA_VISIBLE_DEVICES=0 \
python main_single_gpu.py \
    -cfg='./configs/repmlpres50_light_224_train.yaml' \
    -dataset='imagenet2012' \
    -batch_size=128 \
    -data_path='/dataset/imagenet' \
    -eval \
    -pretrained='./RepMLP-Res50-light-224_train'

Run evaluation using multi-GPUs:

sh run_eval_multi.sh

or

CUDA_VISIBLE_DEVICES=4,5,6,7 \
python main_multi_gpu.py \
    -cfg='./configs/repmlpres50_light_224_train.yaml' \
    -dataset='imagenet2012' \
    -batch_size=16 \
    -data_path='/dataset/imagenet' \
    -eval \
    -pretrained='./RepMLP-Res50-light-224_train'

Training

To train the ResMLP Transformer model on ImageNet2012 with single GPUs, run the following script using command line:

sh run_train.sh

or

CUDA_VISIBLE_DEVICES=0 \
python main_single_gpu.py \
    -cfg='./configs/repmlpres50_light_224_train.yaml' \
    -dataset='imagenet2012' \
    -batch_size=32 \
    -data_path='/dataset/imagenet' \

Run training using multi-GPUs:

sh run_train_multi.sh

or

CUDA_VISIBLE_DEVICES=0,1,2,3 \
python main_multi_gpu.py \
    -cfg='./configs/repmlpres50_light_224_train.yaml' \
    -dataset='imagenet2012' \
    -batch_size=16 \
    -data_path='/dataset/imagenet' \ 

Visualization Attention Map

(coming soon)

Reference

@article{ding2021repmlp,
title={RepMLP: Re-parameterizing Convolutions into Fully-connected Layers for Image Recognition},
author={Ding, Xiaohan and Xia, Chunlong and Zhang, Xiangyu and Chu, Xiaojie and Han, Jungong and Ding, Guiguang},
journal={arXiv preprint arXiv:2105.01883},
year={2021}
}@article{melaskyriazi2021doyoueven,
  title={Do You Even Need Attention? A Stack of Feed-Forward Layers Does Surprisingly Well on ImageNet},
  author={Luke Melas-Kyriazi},
  journal=arxiv,
  year=2021
}