Fix mask generation in base_loader when reading an image

[rod409]

This pull request proposes a bug fix. In base_loader.py, the read_images function generates a mask without checking keys in an original copy. This causes problems on some datasets where mask values are incorrectly replaced. The change would make the mask generation in the base_loader.py match the method used in uniform.py.

semantic-segmentation/datasets/base_loader.py

Lines 176 to 178 in 7726b14

	mask = mask.copy()
	for k, v in self.id_to_trainid.items():
	binary_mask = (mask == k) #+ (gtCoarse == k)

semantic-segmentation/datasets/uniform.py

Lines 112 to 115 in 7726b14

	mask_copy = mask.copy()
	if id2trainid:
	for k, v in id2trainid.items():
	binary_mask = (mask_copy == k)