FFNSR @ NTIRE 2020
This repository is Pytorch code for our proposed FFNSR.
The code is developed by team sysu-AIR, and tested on Ubuntu 18.04 environment (Python 3.6, PyTorch 1.4.0, CUDA 10.1) with 2080Ti GPUs.The details about our proposed FFNSR can be found in our factsheet.
The sysu-AIR team proposed A Fast Feedback Network for Large Scale Image Super-Resolution. Inspired by SRFBN and IMDN, the proposed FFNSR is still reserved the RNN structure but with a information multi-distillation module (IMDM), which can benefit image SR tasks and accelerate inference speed.
-
Number of parameters: 2,099,625
-
Average PSNR on validation data: 29.01096 dB
-
Average inference time (RTX 2080 Ti) on validation data: 4.35 second
Note: We selected the best average inference time among three trials
- Python 3 (Anaconda is recommended)
- skimage
- imageio
- Pytorch (Pytorch version >=0.4.1 is recommended)
- tqdm
- pandas
- cv2 (pip install opencv-python)
-
Clone this repository:
git clone https://github.com/jzrita/NTIRE2020_sysu-AIR.git
-
Download our pre-trained model from the links below, unzip the models and place them to
./models
.Click_here_to_download (code: a3mu)
-
CD the folder and install the requirements:
cd NTIRE2020_sysu-AIR && pip install -r requirements.txt
-
Place the LR pictures to
./picture
../picture/1601.png ./picture/1602.png ./picture/1603.png ...
-
Edit
./options/test/test_SRFBN_example.json
for your needs according to./options/test/README.md
. -
Then run the following commands to test the model:
python test.py
-
Edit
./options/train/train_SRFBN.json
for your needs according to./options/train/README.md
. -
Run command to train the model:
cd NTIRE2020_sysu-AIR python train.py
-
You can monitor the training process in
./experiments
.
-
Download and view our test result.
Click_here_to_download (code: fiz9)