Adjust adjust_* transforms #3222
Conversation
|
@voldemortX Thanks a lot for your contributions, this is really helpful. I'll review the PR shortly. Given #3227, do you think it's worth looking on the other methods to see if there other things we missed? It's totally up to you obviously because it depends on your bandwidth. Let me know. :) |
@datumbox Well, #3227 is already found and marked in the current doc. Also this time I checked again for other transforms to make sure nothing is missed in the new documentation. However, I mostly did everything by looking at the codes. Maybe if I do some actual testing I can find some hidden things? Or maybe we need another set of eyes on this thing just to be sure. I do have some time every now and then, and I'd like to continue working on these things, for instance the Pad fill problem in #3227. |
There was a problem hiding this comment.
@voldemortX LGTM!
Checking the code is the best option especially since you are very familiar with the specific part of the codebase. Thanks a lot for your contribution.
Summary: * adjust_hue * adjust_* * colorjitter Reviewed By: datumbox Differential Revision: D25954551 fbshipit-source-id: 5c849f6f49b92df0d07f73cf507f9db11ad0906e
PR for #3219, I checked the adjust_* functions (I count 6 of them) and corresponding random transform classes (count 2).
Other than
adjust_hue(), I only findadjust_gamma()does not support alpha channel. Others seem to have considered transparency with PIL images.There is one other thing I need to check for sure though: 1-channel and 3-channel exactly represents grayscale and RGB for tensor images right?
cc @datumbox