RepVGG¶

RepVGG: Making VGG-style ConvNets Great Again

Introduction¶

RepVGG, a vgg-style architecture that outperforms many complex models

Its main highlights are:

The model has a normal (a.k.a. feedforward) structure like vgg, without any other branches, each layer takes the output of its only previous layer as input, and feeds the output to its only next layer.
The body of the model uses only 3 × 3 conv and ReLU.
The specific architecture (including specific depth and layer width) is instantiated without automatic search, manual refinement, compound scaling, and other complicated designs.

				Pynative	Pynative	Graph	Graph
	Model	Top-1 (%)	Top-5 (%)	train (s/epoch)	Infer (ms)	train(s/epoch)	Infer (ms)	Download	Config
GPU	RepVGG_A0	71.98	90.36					model	config
Ascend	RepVGG_A0	71.87	90.43

The yaml config files that yield competitive results on ImageNet for different models are listed in the configs folder. To trigger training using preset yaml config.
```
comming soon
```
Here is the example for finetuning a pretrained RepVGG_A0 on CIFAR10 dataset using Adam optimizer.
```
python train.py --model=RepVGG_A0 --pretrained --opt=adam --lr=0.001 ataset=cifar10 --num_classes=10 --dataset_download
```
Detailed adjustable parameters and their default value can be seen in config.py

To validate the model, you can use validate.py. Here is an example to verify the accuracy of pretrained weights.
```
python validate.py --model=RepVGG_A0 --dataset=imagenet --val_split=val --pretrained
```

To validate the model, you can use validate.py. Here is an example to verify the accuracy of your training.

python validate.py --model=RepVGG_A0 --dataset=imagenet --val_split=val --ckpt_path='./ckpt/RepVGG_A0-best.ckpt' 