search

shahaf1313 / ProCST

Official implementation of ProCST image-to-image translation for UDA-SS.
MIT License
29 stars 3 forks source link

TrainID of GTA5 #3

Closed yangH8 closed 2 years ago

yangH8 commented 2 years ago

How to convert the TrainID of GTA5 ?

shahaf1313 commented 2 years ago

Hey Yang, ProCST's translation process translates only images and preserve their content, thus no need to translate the labels. Moreover, in the training process ProCST uses the original labels of GTA5, thus no need to convert them also before training.

yangH8 commented 2 years ago

Thank you for your reply. How to translates the image? Which program to run?

shahaf1313 commented 2 years ago

Please refer to the README.md file for detailed information on both the training and evaluation (translation) of the model. The explanation covers all the prerequisites and describes what scripts you should use in each step.

yangH8 commented 2 years ago

The translation of GTA5 images is not mentioned in README.md.

shahaf1313 commented 2 years ago

Please refer to the "Boosting UDA Methods" section in the README.md file. In this section we describe how to translate images ("boost performance") from both GTA5 and Synthia datasets: image

yangH8 commented 2 years ago

Nice work! If I don't use boosting UDA methods, how can I process GTA5 datasets. Just input raw images?

yangH8 commented 2 years ago

we run the program in training.py. for batch_num, ((source_scales, source_label), target_scales) in enumerate(zip(opt.source_loaders[opt.curr_scale], opt.target_loaders[opt.curr_scale])): RuntimeError: stack expects each tensor to be equal size, but got [127957, 1] at entry 0 and [127088, 1] at entry 1.

I have set the the same number of the input for source and target data

shahaf1313 commented 2 years ago

You shouldn't run training.py. This file is wrapped with other scripts. If you want to train the model, refer to the "Train ProCST Model" section in the README file: image