Skip to content

Latest commit

 

History

History
64 lines (41 loc) · 1.98 KB

edsr.md

File metadata and controls

64 lines (41 loc) · 1.98 KB

Quantization for Low-level Super Resolution

The repo https://github.com/blueardour/EDSR-PyTorch is still marked privated. Refer to https://github.com/billhhh/FQSR if needed.

Install

  1. install quantization dependent packages according to classification.md

    Also, refer the original dependencies for the SR task dependencies.

  2. download the quantization version of EDSR-PyTorch project.

    cd /workspace/git/
git clone https://github.com/blueardour/EDSR-PyTorch
# checkout the quantization branch
cd EDSR-PyTorch
git checkout quantization

  3. make sure the symbolic link is correct. Create the link if it does not exist.

    cd /workspace/git/EDSR-PyTorch/src/
ls -l third_party
# the third_party/quantization should point to /workspace/git/model-quantization

Dataset / Train / Test

Refer instruction from EDSR-PyTorch: how-to-train-edsr-and-mdsr

Link your training data to /workspace/git/EDSR-PyTorch/data The train.sh script in src folder finds data in ../data folder by default.

To start

cd /workspace/git/EDSR-PyTorch/src
bash train.sh config.*

Ongoing config.* is advised to put in /workspace/git/EDSR-PyTorch/src. Verified files which are to be released could be moved to /workspace/git/EDSR-PyTorch/src/config.

Pretrained models and quantization results

We provide pretrained models gradually in google drive

Cite

If you found the work useful, please cite

@misc{wang2021fully,
      title={Fully Quantized Image Super-Resolution Networks}, 
      author={Hu Wang and Peng Chen and Bohan Zhuang and Chunhua Shen},
      year={2021},
      booktitle=ACMMM,
}