Repvgg: Making vgg-style convnets great again¶
Abstract¶
We present a simple but powerful architecture of convolutional neural network, which has a VGG-like inference-time body composed of nothing but a stack of 3x3 convolution and ReLU, while the training-time model has a multi-branch topology. Such decoupling of the training-time and inference-time architecture is realized by a structural re-parameterization technique so that the model is named RepVGG. On ImageNet, RepVGG reaches over 80% top-1 accuracy, which is the first time for a plain model, to the best of our knowledge. On NVIDIA 1080Ti GPU, RepVGG models run 83% faster than ResNet-50 or 101% faster than ResNet-101 with higher accuracy and show favorable accuracy-speed trade-off compared to the state-of-the-art models like EfficientNet and RegNet.
Citation¶
@inproceedings{ding2021repvgg,
title={Repvgg: Making vgg-style convnets great again},
author={Ding, Xiaohan and Zhang, Xiangyu and Ma, Ningning and Han, Jungong and Ding, Guiguang and Sun, Jian},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={13733--13742},
year={2021}
}
Pretrain model¶
Model | Epochs | Params(M) | Flops(G) | Top-1 (%) | Top-5 (%) | Config | Download |
---|---|---|---|---|---|---|---|
RepVGG-A0* | 120 | 9.11(train) | 8.31 (deploy) | 1.52 (train) | 1.36 (deploy) | 72.41 | 90.50 | config (train) | config (deploy) | model |
RepVGG-A1* | 120 | 14.09 (train) | 12.79 (deploy) | 2.64 (train) | 2.37 (deploy) | 74.47 | 91.85 | config (train) | config (deploy) | model |
RepVGG-A2* | 120 | 28.21 (train) | 25.5 (deploy) | 5.7 (train) | 5.12 (deploy) | 76.48 | 93.01 | config (train) |config (deploy) | model |
RepVGG-B0* | 120 | 15.82 (train) | 14.34 (deploy) | 3.42 (train) | 3.06 (deploy) | 75.14 | 92.42 | config (train) |config (deploy) | model |
RepVGG-B1* | 120 | 57.42 (train) | 51.83 (deploy) | 13.16 (train) | 11.82 (deploy) | 78.37 | 94.11 | config (train) |config (deploy) | model |
RepVGG-B1g2* | 120 | 45.78 (train) | 41.36 (deploy) | 9.82 (train) | 8.82 (deploy) | 77.79 | 93.88 | config (train) |config (deploy) | model |
RepVGG-B1g4* | 120 | 39.97 (train) | 36.13 (deploy) | 8.15 (train) | 7.32 (deploy) | 77.58 | 93.84 | config (train) |config (deploy) | model |
RepVGG-B2* | 120 | 89.02 (train) | 80.32 (deploy) | 20.46 (train) | 18.39 (deploy) | 78.78 | 94.42 | config (train) |config (deploy) | model |
RepVGG-B2g4* | 200 | 61.76 (train) | 55.78 (deploy) | 12.63 (train) | 11.34 (deploy) | 79.38 | 94.68 | config (train) |config (deploy) | model |
RepVGG-B3* | 200 | 123.09 (train) | 110.96 (deploy) | 29.17 (train) | 26.22 (deploy) | 80.52 | 95.26 | config (train) |config (deploy) | model |
RepVGG-B3g4* | 200 | 83.83 (train) | 75.63 (deploy) | 17.9 (train) | 16.08 (deploy) | 80.22 | 95.10 | config (train) |config (deploy) | model |
RepVGG-D2se* | 200 | 133.33 (train) | 120.39 (deploy) | 36.56 (train) | 32.85 (deploy) | 81.81 | 95.94 | config (train) |config (deploy) | model |
Models with * are converted from other repos.
Reparameterize RepVGG¶
The checkpoints provided are all in train
form. Use the reparameterize tool to switch them to more efficient deploy
form, which not only has fewer parameters but also less calculations.
python ./tools/convert_models/reparameterize_repvgg.py ${CFG_PATH} ${SRC_CKPT_PATH} ${TARGET_CKPT_PATH}
${CFG_PATH}
is the config file, ${SRC_CKPT_PATH}
is the source chenpoint file, ${TARGET_CKPT_PATH}
is the target deploy weight file path.
To use reparameterized repvgg weight, the config file must switch to the deploy config files as below:
python ./tools/test.py ${RapVGG_Deploy_CFG} ${CHECK_POINT}