QuickSRNet Medium 2x

Click a block to read its description, or select it with Tab and Enter.

QuickSRNet Medium 2xRGB super-resolution, native eval. Input 3 × 64 × 64. All shapes exclude batch.LibreYOLOQuickSRNet Medium 2xRGB super-resolution, native eval. Input 3 × 64 × 64. All shapes exclude batch.Super-resolution networkInput RGB3 × 64 × 64Conv2d 3×3, stride 13 to 32 channels; padding 1Hardtanh [0,1]32 × 64 × 64Feature block, n=532 × 64 × 64Conv2d 3×3, stride 132 to 12 channels; padding 1Hardtanh [0,1]12 × 64 × 64PixelShuffle ×2Rearrange 12 channels to 3 RGB channelsRestored RGB3 × 128 × 128Repeated feature blockFeature input32 × 64 × 64Conv2d 3×332 to 32; stride 1, padding 1, bias=TrueHardtanh [0,1]Elementwise clamp to [0,1]Feature output32 × 64 × 64PixelShuffle indexing12 channels at 64 × 64Each RGB channel has four subpixel positionsReshape and permute3 × 2 × 2 × 64 × 64RGB at 128 × 128Channel positions become spatial offsetsSource: libreyolo/models/quicksrnet/nn.py and model.py. Revision a4d0ecc9e17f.libreyolo.com