Fix RTDetrHungarianMatcher crash on infinite cost matrix#47016
Open
AyushDas4890 wants to merge 6 commits into
Open
Fix RTDetrHungarianMatcher crash on infinite cost matrix#47016AyushDas4890 wants to merge 6 commits into
AyushDas4890 wants to merge 6 commits into
Conversation
4 tasks
Contributor
|
[For maintainers] Suggested jobs to run (before merge) run-slow: rt_detr |
Contributor
CI recapDashboard: View test results in Grafana |
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What does this PR do?
Fixes #47000
RTDetrHungarianMatchercrashes withValueError: cost matrix is infeasiblewhen the cost matrix contains infinite values. Under AMP, fp16 sigmoid
saturation gives probabilities of exactly 0.0 or 1.0, and the 1e-8 epsilon in
the focal class cost underflows in fp16, so the log() terms overflow to ±inf
before reaching
scipy.optimize.linear_sum_assignment.This applies the same guard that was merged for Mask2Former in #27897:
clamp the cost matrix to [-1e10, 1e10] before the assignment.
DFineLossreuses this matcher, so D-FINE is fixed too.
Added a regression test that reproduces the failure with fp16 saturating
logits (both directions) and verifies the matcher returns valid indices.
The test fails without the fix and passes with it.
Disclosure: I used an AI assistant to help diagnose the root cause and draft
this fix myself; I have reviewed the changes and run the new test myself.