isl-org/PhotorealismEnhancement

Is this repo still mantained???? can anyone run this repo without any issue?

lm17918 opened this issue · 4 comments

lm17918 commented
Is this repo still mantained???? can anyone run this repo without any issue?
stefanos50 commented

Hi, I don't know if it is still maintained, but if you have the proper data, then you can train the model and test it in offline frames without any issues. For real-time inference, you need to integrate it on your own.

lm17918 commented

@stefanos50 Are you able to run it from zero now? Can you just give us the folder structure and some sample data we can use (Dummy data not from Cityscapes)? The instructions for creating the dataset are unclear and you can see it if from the periodic issues that are open about this problem.

rogerfinnerty commented

For training you need these files:

  1. fake dataset (I'm using CARLA)
    train/test/val.txt --> each line has image (png or jpg), robust labels - I used MSEG (png or jpg), gbuffers (.npz), and ground truth labels (png or jpg); all files separated by a comma

  2. real dataset (I'm using NuScenes)
    files.txt --> each line has image and robust labels (MSEG), separated by comma

Then just follow the readme to create the matched crops and crop weights files and modify train_pfd2cs.yaml with your file paths and you can start training.

BasuPrithviraj commented

@rogerfinnerty Hi, can you please tell me if you had faced any difficulties with regards to discriminators.py?