Proposal
random_flip_up_down currently accepts only key, image, and probability, while the deterministic flip_up_down function accepts channel_axis for both HWC and CHW images.
Could the random wrapper expose the same keyword argument and forward it to flip_up_down, as was done for random_flip_left_right? This would keep the random and deterministic augmentation APIs consistent and make channel-first pipelines easier to use.
A small regression test covering a CHW image would also help prevent the two APIs from drifting again.
Proposal
random_flip_up_downcurrently accepts onlykey,image, andprobability, while the deterministicflip_up_downfunction acceptschannel_axisfor both HWC and CHW images.Could the random wrapper expose the same keyword argument and forward it to
flip_up_down, as was done forrandom_flip_left_right? This would keep the random and deterministic augmentation APIs consistent and make channel-first pipelines easier to use.A small regression test covering a CHW image would also help prevent the two APIs from drifting again.