slchenchn / PolSAR_N2N

neighbor2neighbor in PolSAR

neighbor2neigbor for PolSAR data

folder

deprecated: erroneous code, which is deprecated, but still valuable

note

model: unetpp.py,

now

• without model summary
• without printing cfg to tensorboard
• without wrapped loss func
• the expectation value of logarithm of SAR intensity is not equal to logarithm of its expectation vallue, correction is needed. See MuLoG, assume number of looks = 1
• turn off the check of assigned maximum should be less than real max value: /home/csl/anaconda3/lib/python3.7/site-packages/piq/utils/common.py", line 54, and set the allow_negative to True

TODO

• rand_pool
• test the PolSAR data loader
• write the N2N loss
• debug the model, the loss value is too large, either log the Hoekman data or somethings else
• run multiple trails using .sh file 2021-05-26
• accelarate the data loading time
• revise PSNR and SSIM, discard data_range param
• write inverse hoekman decomposition, show filtering result