Add OV model for keypoint detection #3957
Conversation
* Update test_cli.py * Update tests/e2e/cli/test_cli.py Co-authored-by: Eunwoo Shin <eunwoo.shin@intel.com> * Update test_cli.py * Update test_cli.py --------- Co-authored-by: Eunwoo Shin <eunwoo.shin@intel.com>
Remove init_args
Remove init_args
Update docs
fix mean and scale
* Bump MAPI * Update exportable code requirements
…#3891) * change drop pixels value * go safe, change only tested models * minor
* Update step size for eff v2 * Update effb0 recipe
update maskrcnn swimt model type to transformer
…form#3903) * add rtmpose_tiny for single obj * add rtmpose_tiny for single obj * modify test subset name * fix unit test * update recipe with reset
…m#3907) * Add warmup_iters for effv2 * Update max_epochs
* Fix pretrained_weight for timm * Fix unit-test
* add rtmpose_tiny for single obj * modify test subset name * fix unit test * property for pck
* Fix cached dir * Pretrained weight download unit-test * Fix pre-commit
…m#3920) Fix datamodule for ov
fix names for ignored scope
* Restore output_raw_scores for classificaiton * Add uts * Fix linter
add label info to init Signed-off-by: Ashwin Vaidya <ashwinnitinvaidya@gmail.com>
* Fix binary classification * Add unit-tests
* ignore heads and disable smooth quant * add activations_range_estimator_params * update changelog
…3931) * Fix Task Chain * Add multi-label case as well * Add multi-label case as well2 * Add H-label case
…ies using label's id/key (open-edge-platform#3932) Modify label_groups for dm_label_categories with id/key of label
…-platform#3933) * remove datumaro attribute id from tiling * add subset names
github-actions
bot
added
DEPENDENCY
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## releases/2.2.0 #3957 +/- ##
=================================================
Coverage ? 80.79%
=================================================
Files ? 276
Lines ? 27302
Branches ? 0
=================================================
Hits ? 22059
Misses ? 5243
Partials ? 0
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
|
|
||
| kpts_visible_stacked = np.stack(kpts_visible) | ||
|
|
||
| normalize = np.tile(np.array([self.input_size[::-1]]), (pred_kpts.shape[0], 1)) |
There was a problem hiding this comment.
the reason why input size order is reversed is that norm_factor order is (w, h)? Referring docstring of the function, it should be (h, w). Which one is right?
There was a problem hiding this comment.
It's wh, since we use xy keypoints coordinates. The function is order-agnostic, we can use it for both xy and xy. The main point is that we need to pass aligned kps and normalization, which are in xy format for our case.
There was a problem hiding this comment.
Thanks for explanation. Could you update norm_factor docstring at keypoint_pck_accuracy then?
|
@sovrasov - Could you please designate the target branch as the develop branch? OTX2.2.0 has already proven that KeyPoint detection supports train and export. OV IR inference validation will be available in the upcoming OTX2.3.0. |
|
@chuneuny-emily just in case, export of keypoint model in 2.2 can be launched, but it generates an incorrect model, so it can be used only as a mock. |
Add integration tests for keypoint detection task (open-edge-platform#3945) * Fix export * Add integrastion tests for keypoint detection * Update export * Add kp det to gh workflow * Fix ruff * Move std mean to class vars Cleanup Implement dummy input for kp Update export inf test Upgrade MAPI Cleanup Fix CLI tests Cleanup Fix kp OV model data config Fix dummy input docstring Co-authored-by: Prokofiev Kirill <kirill.prokofiev@intel.com> Add unit tests for kp det model Fix ruff Update pck docs Fix typos in tests Add uts for PCK metric Rename kaypoints det metric Del extra transform in p ove model Fix unit test naming Fix cpu train/eval Fix autocast deprecation warnings
11 participants
Summary
Results on coco_keypoint_single_obj/medium: