Closed wenliangzhao2018 closed 11 months ago
This pull request was exported from Phabricator. Differential Revision: D47682434
This pull request was exported from Phabricator. Differential Revision: D47682434
This pull request was exported from Phabricator. Differential Revision: D47682434
This pull request was exported from Phabricator. Differential Revision: D47682434
Summary: X-link: https://github.com/facebookresearch/detectron2/pull/5048
Unet support different aspect ratio by natural. For training we need to make change to group images with similar aspect ratio into the same batch and do padding/cropping. For inference, we need to change start_code dimension accordingly.
Here is a result showing 1:1 and 16:9 inference are similar. It supports the mixed training decision. N4024922
Differential Revision: D47682434