QuickSRNet Medium 2x
Fit diagram
Read at 100%
Clear selection
Download SVG
Download PNG
Click a block to read its description, or select it with Tab and Enter.
QuickSRNet Medium 2x
RGB super-resolution, native eval. Input 3 × 64 × 64. All shapes exclude batch.
LibreYOLO
QuickSRNet Medium 2x
RGB super-resolution, native eval. Input 3 × 64 × 64. All shapes exclude batch.
Super-resolution network
Input RGB
3 × 64 × 64
Conv2d 3×3, stride 1
3 to 32 channels; padding 1
Hardtanh [0,1]
32 × 64 × 64
Feature block, n=5
32 × 64 × 64
Conv2d 3×3, stride 1
32 to 12 channels; padding 1
Hardtanh [0,1]
12 × 64 × 64
PixelShuffle ×2
Rearrange 12 channels to 3 RGB channels
Restored RGB
3 × 128 × 128
Repeated feature block
Feature input
32 × 64 × 64
Conv2d 3×3
32 to 32; stride 1, padding 1, bias=True
Hardtanh [0,1]
Elementwise clamp to [0,1]
Feature output
32 × 64 × 64
PixelShuffle indexing
12 channels at 64 × 64
Each RGB channel has four subpixel positions
Reshape and permute
3 × 2 × 2 × 64 × 64
RGB at 128 × 128
Channel positions become spatial offsets
Source: libreyolo/models/quicksrnet/nn.py and model.py. Revision a4d0ecc9e17f.
libreyolo.com