- main framework: tensorflow: 1.2 or above
- matlab python api
matlabroot is the matlab root path.
cd "matlabroot/extern/engines/python"
python setup.py install
then we load eval matlab commands in py env:
>>> import matlab
>>> matlab.__path__
['/usr/local/lib/python2.7/dist-packages/matlab']
references: https://cn.mathworks.com/help/matlab/matlab_external/install-the-matlab-engine-for-python.html
CUDA_VISIBLE_DEVICES=0 python test.py --gpu_id=0 --batch_size=1 --channel=1 --filter_num=64 --scale=4 --sr_method=EDSR_v313 --model_path=./ckpt/EDSR_v313/EDSR_v313-epoch-1-step-20000-2018-02-12-02-56.ckpt-20000 --image=./demo/test.png --output_dir=./ --scale=4
CUDA_VISIBLE_DEVICES=3 python solver.py --gpu_id=3 --dataset_dir=./dataset/div2k_difficulty_x50.h5 --g_log_dir=./log/LapSRN_v7 --g_ckpt_dir=./ckpt/LapSRN_v7 --default_sr_method='LapSRN_v7' --test_dataset_path=./dataset/mat_test/set5/mat --epoches=1 --inner_epoches=1 --default_channel=1 --upscale_factor=4 --filter_num=64 --batch_size=2
the image dataset and models are uploaded on baiduyun
update the model_path, model_name, img_dir, output_dir in ntire_test.py file:
model_path = 'ckpt/EDSR_v313/EDSR_v313-epoch-1-step-20000-2018-02-12-02-56.ckpt-20000'
model_name = 'EDSR_v313'
img_dir = '../dataset/DIV2K_valid_LR_difficult'
output_dir =' ../dataset/VALIDATION_V4/res'
then 'python ntire_test.py' to generate the super-resolution images. current results is
PSNR:21.744125
SSIM:0.460918
generate the image the test:
CUDA_VISIBLE_DEVICES=0 python val.py --gpu_id=0 --channel=1 --filter_num=64 --sr_method=LapSRN_v7 --model=./ckpt/LapSRN_v7/LapSRN_v7-epoch-2-step-9774-2017-07-23-13-59.ckpt-9774 --image=./dataset/mat_test/set5/mat --scale=4
the (pnsr/ssim/ifc) results on set5/set14/bsd100 results:
| method | upscale_factor | set5 | set14 | bsd100 |
|---|---|---|---|---|
| SRCNN | 4 | 30.49/0.862/2.997 | 27.61/0.754/2.767 | 26.91/0.712/2.412 |
| DRRN | 4 | 31.68/0.889/3.702 | 28.31/0.774/3.254 | 27.38/0.728/2.746 |
| LapSRN | 4 | 31.54/0.885/3.559 | 28.19/0.772/3.147 | 27.32/0.728/2.677 |
| ours | 4 | 31.67/0.890/3.620 | 28.26/0.774/3.207 | 27.36/0.729/2.698 |
| SRCNN | 8 | 25.33/0.689/0.938 | 23.85/0.593/0.865 | 24.13/0.565/0.705 |
| LapSRN | 8 | 26.14/0.738/1.302 | 24.44/0.623/1.134 | 24.54/0.586/0.893 |
| ours | 8 | 26.22/0.747/1.340 | 24.58/0.627/1.193 | 24.61/0.588/0.928 |
- Chao Dong, Chen Change Loy, Kaiming He, Xiaoou Tang. Learning a Deep Convolutional Network for Image Super-Resolution, in Proceedings of European Conference on Computer Vision (ECCV), 2014
- Shi W, Caballero J, Huszar F, et al. Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network[C]// CVPR 2016:1874-1883.
- Accurate image super-resolution using very deep convolutional networks[C] //Computer Vision and Pattern Recognition, Las Vegas, 2016: 1646-1654.
- Kim J, Lee J K, Lee K M. Deeply-recursive convolutional network for image super-resolution[C]//Computer Vision and Pattern Recognition, Las Vegas, 2016: 1637-1645.
- Johnson, Justin, Alexandre Alahi, and Li Fei-Fei. "Perceptual losses for real-time style transfer and super-resolution." European Conference on Computer Vision. Springer International Publishing, 2016.
- Wei-Sheng Lai, Jia-Bin Huang, Narendra Ahuja, and Ming-Hsuan Yang, Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution, CVPR, 2017