Expand description
RepVGG inference implementation
Key characteristics:
- Efficient inference architecture through structural reparameterization
- Single 3x3 conv layer after fusing 3x3 branch, 1x1 branch and identity branch
- Different configurations including a0-a2, b0-b3 and variants with group convolutions
- High accuracy with VGG-like plain architecture and training
References:
- RepVGG Paper. RepVGG: Making VGG-style ConvNets Great Again
- Official Implementation