Issues in fisher score computation

[ajeetkverma]

I have used all the deafault settings, getting stucked at this point:

[24/07/20 16:16:14] [srtta.py: 123]: use GaussianBlur for im_5 Traceback (most recent call last): File "C:\Users\akver\Desktop\VSR\SRTTA\src\main_tta.py", line 287, in main(args) File "C:\Users\akver\Desktop\VSR\SRTTA\src\main_tta.py", line 186, in main img_out = srtta_model(train_loader, img_lr, img_gt, filename[0], corruption) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\akver\Desktop\VSR\SRTTA\src\srtta.py", line 125, in call self.test_time_adaptation(train_loader, corruptions) File "C:\Users\akver\Desktop\VSR\SRTTA\src\srtta.py", line 187, in test_time_adaptation self.fisher_restoration() File "C:\Users\akver\Desktop\VSR\SRTTA\src\srtta.py", line 217, in fisher_restoration mask = self.fishers[f"{nm}.{npp}"][-1]

KeyError: 'body.0.body.0.weight'

[Caesarhhh]

It seems that the function compute_fisher has not been executed. Please debug and check if the function is executing normally.

[ajeetkverma]

solved it by commenting line in srtta.py

if t_data.dataset.name != "Set5": continue

i found this line was making code to skip the fishers metric, and causing the issues.

Thank you for prompt reply.