For the basic setting of this work, please prepare datasets of Cityscapes, Stanford2D3D, and DensePASS.
Our proposed DensePASS is available at Google Drive.
The DensePASS dataset has 100 panoramic images and annotations for evaluation.
The other unlabeled images could be found at WildPASS.
'train.py' in S_R, A_S and S_A_R are used for training on outdoor dataset with different modules settings.
The training configurations can be adjusted at 'train.py' or by parameters like --model DANet.
An example of training is 'python train.py'.
This code is heavily borrowed from AdaptSegNet.