Fix accuracy drop in MaskRCNN #3562
Merged
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
eugene123tw
requested review from
samet-akcay,
harimkang,
vinnamkim,
jaegukhyun,
kprokofi,
chuneuny-emily,
sovrasov,
sungchul2,
GalyaZalesskaya,
negvet,
goodsong81,
yunchu,
wonjuleee and
eunwoosh
as code owners
eugene123tw
changed the title
|
@kprokofi do we need to target all PRs to develop considering the new release schedule? |
|
Yes, @eugene123tw please, retarget your PR to develop. CF for 2.0 is already finished. |
sovrasov
approved these changes
May 30, 2024
|
@sovrasov @kprokofi @eugene123tw - Please merge this PR into the releases/2.0.0 branch. CF has already passed, but the E2 defects are still open, and this PR is a PR to fix one of defects. OTX2.0.0 cannot be released if there are E1/E2 defects. |
harimkang
approved these changes
May 31, 2024
harimkang
added a commit
that referenced
this pull request
Jun 19, 2024
* Remove LITMODULE_PER_TASK * Disable Resnext101_ATSS model on XPU (#3514) * diable resnext101_atss on XPU * give detailed xpu device info to perf tag * revert debug code --------- Co-authored-by: kirill prokofiev <kirill.prokofiev@intel.com> * Remove invalid perf benchmark reference history for v2.0.0 (#3517) * Fix a bug that dino_v2 model can't be run w/ HPO (#3518) * add reduce function to dino backbone * add unit test * update integration test * change name * Fix detection export performance degradation (#3520) * Fix to use right index * Align forward by adding missed parts * Revert metrics threshold (#3528) Revert threshold Signed-off-by: Ashwin Vaidya <ashwinnitinvaidya@gmail.com> * Tile with full img optional (#3530) * make tile dataset with full image optional * Add a feature to adapt max value of HPO batch size search space (#3532) * implement adaptive max value of bs search space * implement unit test * Remove duplicates in get_idx_list_per_classes (#3537) * Remove duplicates in get_idx_list_per_classes * Reduce time complex * Revisit Docker image build script (#3536) * Revisit Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Fix test Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> --------- Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Bump datumaro version to 1.6.1 (#3535) * Add F1 metric computation during detection tasks (#3539) * add f1 metric during training for detection * add f1 metric during training for detection * add a new metric * remove configure_metric in det model * Update doctstring for MeanAveragePrecisionFMeasure Co-authored-by: Vinnam Kim <vinnam.kim@gmail.com> * Trigger Build * resolve precommit error --------- Co-authored-by: Vinnam Kim <vinnam.kim@gmail.com> * Fix yolox export perf degradation (#3534) * Add `Focus` export pipeline * Update `update_ov_subset_pipeline` to update `image_color_channel` * Fix MaskRCNN IR Accuracy Drop (#3540) fix ir maskrcnn accuracy drop * Hotfix/geti integration (#3543) * Fix Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Fix Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> --------- Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Fix Anomaly OV export flag (#3558) swap_rgb to False Signed-off-by: Ashwin Vaidya <ashwinnitinvaidya@gmail.com> * Fix accuracy drop in MaskRCNN (#3562) * fix maskrcnn accuracy drop * pad to square * remove unnecessary changes * remove unnecessary changes * Update codeql workflow to generate a report (#3559) * Fix & Refine HPO (#3565) * refine engine/hpo * implement draft test code * change replace to shutil.copy * use same initial weight during HPO * implement unit test * search_space support Path * move Path out of TYPE_CHECKING * Fix Optimize in Anomaly Task (#3561) * use val_dataloader Signed-off-by: Ashwin Vaidya <ashwinnitinvaidya@gmail.com> * Apply suggestions from code review Co-authored-by: Harim Kang <harim.kang@intel.com> --------- Signed-off-by: Ashwin Vaidya <ashwinnitinvaidya@gmail.com> Co-authored-by: Harim Kang <harim.kang@intel.com> * Update dependencies (#3570) * Bump anomalib from 1.0.1 to 1.1.0 (#3572) Bump anomalib version to 1.1.0 * Fix F1 instance seg accuracy drop (#3578) * add MaskRLEMeanAPFMeasureCallable * update recipes * format * Update for using locally stored pretrained weights in the CI for torch hub models (#3541) * apply changed arg of OTXModel.export() to YOLOX model * update dinov2 pretrained weights loading * skip yolox-tiny-tile xai test * make semi-sl dino support pickle --------- Signed-off-by: Ashwin Vaidya <ashwinnitinvaidya@gmail.com> Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> Co-authored-by: Kang, Harim <harim.kang@intel.com> Co-authored-by: Eunwoo Shin <eunwoo.shin@intel.com> Co-authored-by: kirill prokofiev <kirill.prokofiev@intel.com> Co-authored-by: Kim, Sungchul <sungchul.kim@intel.com> Co-authored-by: Ashwin Vaidya <ashwin.vaidya@intel.com> Co-authored-by: Eugene Liu <eugene.liu@intel.com> Co-authored-by: Vinnam Kim <vinnam.kim@intel.com> Co-authored-by: Wonju Lee <wonju.lee@intel.com> Co-authored-by: Vinnam Kim <vinnam.kim@gmail.com>
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.
Summary
This PR aligns the image padding method between OpenVINO and PyTorch for MaskRCNN. I suggest we verify the detection results, but I currently don't have the bandwidth to do so.
Wgisd_small/1 test result
PyTorch │ test/map_50 │ 0.4580044150352478 │
ov │ test/map_50 │ 0.45944005250930786 │
How to test
Checklist
License
Feel free to contact the maintainers if that's a concern.