HPINet
HPINet copied to clipboard
passerer
Hierarchical Pixel Integration for Lightweight Image Super-Resolution
From Coarse to Fine: Hierarchical Pixel Integration for Lightweight Image Super-Resolution
This is the Pytorch implementation of HPINet (AAAI2023 Acceptance).
[arXiv paper] [pretrained models] [visual results (code:z86e)]
Dependencies
- python == 3.6
- pytorch == 1.8.0
- scikit-image == 0.17.2
- einops == 0.3.2
Other versions of the packages may also work, but they are not tested.
Test Datasets
- Download the five test datasets (Set5, Set14, B100, Urban100, Manga109) from Google Drive
- Put them in
benchmarks/directory, or customize the path in the code
Run Test
python test.py --model L --scale 4
The pretrained models are provided in checkpoints/:
| x2 | x3 | x4 | |
|---|---|---|---|
| S | ✔ | ||
| M | ✔ | ✔ | ✔ |
| L | ✔ |
Train Datasets
DIV2K
├── DIV2K_train_HR
├── DIV2K_train_LR_bicubic
│ ├── X2
│ ├── X3
│ ├── X4
Run Train
Here is an example:
CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7" python train.py --model M --root DIV2K/ --ext .png --scale 4
passerer