DingXiaoH/RepVGG

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

3478 stars · 433 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

RepVGG is a PyTorch implementation of the VGG-style ConvNet architecture from the CVPR 2021 paper, achieving over 84% top-1 ImageNet accuracy via structural re-parameterization. It includes pretrained models, training code, train-to-inference model conversion, and a semantic segmentation example.

Use cases

• train a fast VGG-style image classification model on ImageNet
• convert a multi-branch training model into a deployable single-branch inference model
• use pretrained RepVGG weights as a backbone for semantic segmentation
• benchmark ConvNets against visual transformers like Swin
• integrate re-parameterized backbones into detection models like YOLOv6/YOLOv7

When to choose

• you need high-throughput image classification with simple VGG-like architecture
• you want pretrained ConvNet backbones for downstream vision tasks
• you want to apply structural re-parameterization for faster inference

When to avoid

• you need transformer-based vision models
• you need a framework other than PyTorch without community ports
• you need actively developed features beyond the released models

Facets

library · maturity maintenance

deep-learning machine-learning image-processing computer-vision deep-learning image-processing python pytorch repvgg model-zoo re-parameterization image-classification semantic-segmentation pretrained-models convnet gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="DingXiaoH/RepVGG")