search

lhoyer / DAFormer

[CVPR22] Official Implementation of DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation
Other
466 stars 91 forks source link

using mask in one_mix method in mmseg/models/utils/dacs_transforms.py #85

Open beautifulchoi opened 9 months ago

beautifulchoi commented 9 months ago

hello, first of all, thanks for your great work.

I want to extend this code, but have some question.

in mmseg/models/utils/dacs_transforms.py, there is "one_mix" method. it seems making mixing original image and target image. but, there is only mask[0] used, which is first batch's mask.

but in "get_class_masks" method, the number of generated masks is batch size. why others are not used but only using first mask? image