The key idea of Repvgg is that by using re-parameterization, the model architecture could be trained in multi-branch but validated in single branch. Figure 1 shows the basic model architecture of Repvgg. By utilizing different values for a and b, we could get various repvgg models. Repvgg could achieve better model performance with smaller model parameters on ImageNet-1K dataset compared with previous methods.[1]
Figure 1. Architecture of Repvgg [1]
Our reproduced model performance on ImageNet-1K is reported as follows.
| Model | Context | Top-1 (%) | Top-5 (%) | Params (M) | Recipe | Download |
|---|---|---|---|---|---|---|
| repvgg_a0 | D910x8-G | 72.19 | 90.75 | 9.13 | yaml | weights |
| repvgg_a1 | D910x8-G | 74.19 | 91.89 | 14.12 | yaml | weights |
| repvgg_a2 | D910x8-G | 76.63 | 93.42 | 28.25 | yaml | weights |
| repvgg_b0 | D910x8-G | 74.99 | 92.40 | 15.85 | yaml | weights |
| repvgg_b1 | D910x8-G | 78.81 | 94.37 | 57.48 | yaml | weights |
| repvgg_b2 | D910x64-G | 79.29 | 94.66 | 89.11 | yaml | weights |
| repvgg_b3 | D910x64-G | 80.46 | 95.34 | 123.19 | yaml | weights |
| repvgg_b1g2 | D910x8-G | 78.03 | 94.09 | 45.85 | yaml | weights |
| repvgg_b1g4 | D910x8-G | 77.64 | 94.03 | 40.03 | yaml | weights |
| repvgg_b2g4 | D910x8-G | 78.8 | 94.36 | 61.84 | yaml | weights |
- Context: Training context denoted as {device}x{pieces}-{MS mode}, where mindspore mode can be G - graph mode or F - pynative mode with ms function. For example, D910x8-G is for training on 8 pieces of Ascend 910 NPU using graph mode.
- Top-1 and Top-5: Accuracy reported on the validation set of ImageNet-1K.
Please refer to the installation instruction in MindCV.
Please download the ImageNet-1K dataset for model training and validation.
- Distributed Training
It is easy to reproduce the reported results with the pre-defined training recipe. For distributed training on multiple Ascend 910 devices, please run
# distributed training on multiple GPU/Ascend devices
mpirun -n 8 python train.py --config configs/repvgg/repvgg_a1_ascend.yaml --data_dir /path/to/imagenetIf the script is executed by the root user, the
--allow-run-as-rootparameter must be added tompirun.
Similarly, you can train the model on multiple GPU devices with the above mpirun command.
For detailed illustration of all hyper-parameters, please refer to config.py.
Note: As the global batch size (batch_size x num_devices) is an important hyper-parameter, it is recommended to keep the global batch size unchanged for reproduction or adjust the learning rate linearly to a new global batch size.
- Standalone Training
If you want to train or finetune the model on a smaller dataset without distributed training, please run:
# standalone training on a CPU/GPU/Ascend device
python train.py --config configs/repvgg/repvgg_a1_ascend.yaml --data_dir /path/to/dataset --distribute FalseTo validate the accuracy of the trained model, you can use validate.py and parse the checkpoint path with --ckpt_path.
python validate.py -c configs/repvgg/repvgg_a1_ascend.yaml --data_dir /path/to/imagenet --ckpt_path /path/to/ckpt
Please refer to the deployment tutorial in MindCV.
[1] Ding X, Zhang X, Ma N, et al. Repvgg: Making vgg-style convnets great again[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021: 13733-13742.