search

RyanWri / MaxSR

implementation of MaxSR paper, vision transformer for image super resolution
1 stars 0 forks source link

Embeding + positional encoding #52

Closed RyanWri closed 1 month ago

RyanWri commented 1 month ago
  1. Image patch embedding
  2. Positional Encoding
  3. Components modifications
  4. training efficiency
  5. 0.70 seconds per image
  6. Reconstruction Stitching of 64 patches
  7. Cuda optimizations
  8. Model Checkpoints
  9. Epochs saving
  10. Automatic memory management