Validation of WaveDiff 2.0.0 has been done using the older version of the code in the branch main_random_seed that was upgraded to include the user setting of a fixed random seed.
Tobias' Phase Retrieval implementation in the old code should be upgraded with the fixed random seed in two commit states It's possible to cherry-pick one of the commit states in tf_psf_field , but not the other in long-runs/train_eval_plot_script_click_multi_cycle.py as they include some other unnecessary changes (bad practice on my part). But since they're really simple, so these changes could be added by hand.
Once Tobias updates his Phase Retrieval code (in the old version) it should be used to produce new baseline results to be used for validation.
Validation of WaveDiff 2.0.0 has been done using the older version of the code in the branch main_random_seed that was upgraded to include the user setting of a fixed random seed.
Tobias' Phase Retrieval implementation in the old code should be upgraded with the fixed random seed in two commit states It's possible to cherry-pick one of the commit states in tf_psf_field , but not the other in long-runs/train_eval_plot_script_click_multi_cycle.py as they include some other unnecessary changes (bad practice on my part). But since they're really simple, so these changes could be added by hand.
Once Tobias updates his Phase Retrieval code (in the old version) it should be used to produce new baseline results to be used for validation.