Fixes for pytorch<2.0 in average precision #3356
Closed
Conversation
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
vfdev-5
force-pushed
the
fix-pth-versions-ci
branch
from
e1ff650 to
9444581
Compare
There was a problem hiding this comment.
Pull Request Overview
This PR addresses fixes for PyTorch versions below 2.0 in the computation of average precision and recall for object detection metrics. Key changes include conditional use of the "stable" parameter in torch.argsort based on the torch version, updated device type comparisons (using device.type instead of direct torch.device comparisons), and adapting precision aggregation logic to mitigate type issues on different backends.
Reviewed Changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| ignite/metrics/vision/object_detection_average_precision_recall.py | Adds a torch version check and conditionally passes the "stable" flag to torch.argsort; adjusts precision aggregation logic. |
| ignite/metrics/mean_average_precision.py | Introduces a similar torch version check and updates device type checks for consistency. |
| tests/ignite/metrics/vision/test_object_detection_map.py | Updates device comparison to use device.type for handling the MPS backend. |
Comments suppressed due to low confidence (4)
ignite/metrics/vision/object_detection_average_precision_recall.py:221
- Confirm that the conditional kwargs usage for torch.argsort maintains consistent sorting behavior across different torch versions without performance regressions.
indices = torch.argsort(scores, descending=True, **kwargs)
ignite/metrics/vision/object_detection_average_precision_recall.py:267
- Ensure that the fallback value 0.0 is of the same dtype as precision_integrand to avoid potential type mismatches in the average precision calculation.
precision_integrand = torch.where(
recall_mask,
precision_integrand.take_along_dim(torch.where(recall_mask, rec_thresh_indices, 0), dim=-1),
0.0,
)
ignite/metrics/mean_average_precision.py:348
- The updated device type check using device.type is more robust; verify that this pattern is applied consistently across similar device comparisons.
if tp_summation.device.type != "mps":
tests/ignite/metrics/vision/test_object_detection_map.py:867
- The test now correctly uses device.type for checking the MPS backend; ensure that all MPS-specific skips use this updated checking method.
if device.type == "mps":
vfdev-5
force-pushed
the
fix-pth-versions-ci
branch
from
9444581 to
a7ff713
Compare
This was referenced Jul 7, 2025
|
Failed tests: https://github.com/pytorch/ignite/actions/runs/18508418511/job/52743076044 |
vfdev-5
force-pushed
the
fix-pth-versions-ci
branch
from
4a5e6ca to
71e63c8
Compare
|
Closing as removed 1.X pytorch version support |
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.
Description:
https://github.com/pytorch/ignite/actions/runs/14102362634