-
train_BaggedDIP_speckle.py: training Bagged-DIP based PGD algorithm.
-
function_grad.py: i) explicit gradient function of the MLE loss function, and ii) implementation of Newton Schulz algorithm for efficently approximating matrix inverse.
-
decoder.py: basic network structures of the Deep Image Prior/Deep Decoder we use in Bagged-DIP.
-
utils.py: all the other helper functions.
python train_BaggedDIP_speckle.py
python train_BaggedDIP_speckle.py --compression_rate 0.5 --patch_size 128 --lamb 1.0 --crop True --lr_NN 1e-3 --lr_GD 0.01 --outer_ite 100 --num_look 100 --weight_decay 0.0 --test_name 'Set11' --DIP_avg True --use_complex True
[1] Chen, Xi, Zhewen Hou, Christopher Metzler, Arian Maleki, and Shirin Jalali. "Bagged Deep Image Prior for Recovering Images in the Presence of Speckle Noise." arXiv preprint arXiv:2402.15635 (2024). paper
[2] Chen, Xi, Zhewen Hou, Christopher Metzler, Arian Maleki, and Shirin Jalali. "Multilook compressive sensing in the presence of speckle noise." In NeurIPS 2023 Workshop on Deep Learning and Inverse Problems. 2023. paper
[3] Zhou, Wenda, Shirin Jalali, and Arian Maleki. "Compressed sensing in the presence of speckle noise." IEEE Transactions on Information Theory 68.10 (2022): 6964-6980. paper