[Fix] Flip yaw angle for DepthInstance3DBoxes.overlaps #1548
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Codecov Report
@@ Coverage Diff @@
## dev #1548 +/- ##
==========================================
- Coverage 50.19% 50.15% -0.05%
==========================================
Files 223 223
Lines 19104 19113 +9
Branches 3134 3136 +2
==========================================
- Hits 9590 9586 -4
- Misses 8935 8948 +13
Partials 579 579
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
Merged
Merged
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation
This fix is separated from FCAF3D PR #1547 after a small discussion with @ZwwWayne.
While working on that PR I've checked my FCAF3D checkpoints for ScanNet, SUN RGB-D and S3DIS from
mmdetectiontion3d==0.15.0. And for ScanNet and S3DIS metrics were exactly the same, unlike SUN RGB-D with mAP@0.5 less then expected by ~3%. After some debugging I saw that the model is fine, but the difference is caused by the evaluation function. And basically the difference is inBaseInstance3DBoxes.overlaps. The difference is not that big, i.e.test_indoor_evalsuccessfully passes with this fix and without it. But, maybe it is a bug.As I understand we call
mmcv.ops.box_iou_rotatedinBaseInstance3DBoxes.overlapsand this function is designed for anti-clockwise heading angle. So it is probably ok forLidarInstance3DBoxes. But forDepthInstance3DBoxesthe yaw angle is clockwise, so we need to flip the angle ofboxes1andboxes1before callingmmcv.ops.box_iou_rotated.Also the issue is critical as it has a little affect on
indoor_evaland сonsequently on all indoor models. I have tried the VoteNet checkpoint and the affect of this PR is minimal, e.g. <1% . However looks like VoteNet is 15% mAP@0.5 less than it should be in currentdevbranch. I'm not sure about the reason. May be it should be retrained, or the bug is introduced innms3doriou3drefactoring in last 2 months.Modification
Inherit
DepthInstance3DBoxes.overlapsfromBaseInstance3DBoxes.overlaps.BC-breaking
Yes.