I am testing restormer on the project aberration_sbcho (private) Mostly, it follows 'real-denoising task' of orginal repo. I am not plan to run it on the multi-gpu system.
See original README.md for further informations
- It was submodule but now it is runnable independently
- Made new dataset class for 3D-PSF abrr dataset for generation of dataset instead of on-the-fly use... cuz it was too slow
- prepare to run on remote server, via bash script file : Trained one model with patch size 384...(which causes oom in RTX 4090) It should be merged with local server version...
option file should be passed to the train/test scripts -opt (option)
python basicsr\train.py -opt Denoising\Options\multabrr_mpmneural.yml # for train
python basicsr\test_ch.py -opt Denoising\Options\multabrr_mpmneural_test.yml # for test. should modify config file to pass desired 'resume state'All the config files at Denoising/Options just same as original repo. See README.md for usage...
multabrr_mpmneural.yml: Default, train. should be passed tobasicsr/train.pymultabrr_mpmneural_test.ymlshould bebasicsr/test.pymultabrr_mpmneural_test_AUGdata.ymlshould be passed tobasicsr/test_1stack.pyas option. rename originalvis_outputs,visualizationdirectory on outputs(experiments) directory.- To process dataset
synabrr_mpmneural_3dpsf_example.yml,synabrr_mpmneural_2dpsf_createdset.yml