The network is actually a concise deep unrolling network that utilize the tensor low-rank property of the color image. The network structure is as follows. It consists three parts, X, Z, L. The Z part enforces the low rankness in the CNN-transformed domain. The singular value thresholding (SVT) is utilized to realize the low-rank property.
- the essential requirements is listed in
requirements.txt
. runpip install -r requirements.txt
to install them. - we just run the code in Windows 11 platform with NVIDIA Geforce 1080Ti GPU.
- we also test the code in ubuntu. it's ok to run.
- you could train the network by
python main.py
testmask.py
is just used to generate the undersampling masks for the test images intest_images
folder, e.g.,test_uds_0.1.npz
. you can just ignore it if you don't care the test undersampling mask.- all other info can be found in our paper.