[Question] There are some less convenient settings in transforms.py

[STHxiao]

1.In RandomFlip, RandomRotation, RandomGaussianBlur, and PILRandomGaussianBlur, augmentation is not performed when prob is set to 1, which seems to go against the intuition that the higher the prob, the higher the probability of occurrence. Should the judgment statement be changed to if np.random.rand() > self.prob: return sample_meta 2.In RandomCutOut sample_meta[key][y1: y2, x1: x2, :] = fill_value doesn't apply to the seg_target output from Resize class because it has a shape of (h,w) rather than (h,w,1)