@article{sun2020dual,
title={Dual-path attention network for compressed sensing image reconstruction},
author={Sun, Yubao and Chen, Jiwei and Liu, Qingshan and Liu, Bo and Guo, Guodong},
journal={IEEE Transactions on Image Processing},
volume={29},
pages={9482--9495},
year={2020},
publisher={IEEE}
}
You can use crop.m to generate training data.
You can download the trained model from Google Drive. Then put them in the file folder--checkpoint.
"g_GRAY_1.npz" is only the model of structure-path in the network after first step training. If you want to test the whole network including dual-path, "g_GRAY_2.npz" should be in the checkpoint folder.
$ python DPA-Net.py
$ puthon eval_DPA.py
You can change the setting in config.py, such as MR(measurement ratio) and so on. The files of trained model parameter and train data will be uploaded soon.
Maybe, some warnings or errors will arise when you run the code because of my careless. If you can not cover it, just give me feedback. I will solve it as soon as possible.
phi = np.reshape(A, (block_size, block_size, 1, size_y))
y_meas = tf.nn.conv2d(t_target_image, A, (1, block_size, block_size, 1), padding='SAME')