-
Notifications
You must be signed in to change notification settings - Fork 7.1k
Closed
Description
In our detection references we have a couple of transformations that can handle images and bounding boxes simultaneously:
-
RandomPhotometricDistort
port RandomPhotoMetricDistort to prototype transforms #5663 -
RandomZoomOut
port RandomZoomOut from detection references to prototype transforms #5551 -
RandomIoUCrop
@vfdev-5 [proto] Ported RandomIoUCrop from detection refs #6401 -
ScaleJitter
@pmeier portScaleJitter
from detection reference to prototype transforms #6411 -
FixedSizeCrop
@pmeier portFixedSizeCrop
from detection references to prototype transforms #6417 -
RandomShortestSize
portRandomShortestSize
from detection references to prototype transforms #6418 -
SimpleCopyPaste
@vfdev-5 [proto] Ported SimpleCopyPaste transform #6451
They never made it into torchvision.transforms
since have a different call signature than all other transforms there. Now that torchvision.prototype.transforms
support simultaneous transforming of multiple types, we can migrate them.