Add oriented bounding box head for object detection tasks #73
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
* Added ResNet variants * ResNet50 example * Simplified example * fixed resnet config * [Automated] Updated coverage badge --------- Co-authored-by: Martin Kozlovsky <martin.kozlovsky@luxonis.com> Co-authored-by: GitHub Actions <actions@github.com>
* fixed incorrect class property call * fixed exporter uploading * uploadCheckpoint uploads on every checkpoint epoch * fix temp files names * updated callback readme * pre-commit run
* option to source custom code in CLI * removed empty dicts * [Automated] Updated coverage badge --------- Co-authored-by: GitHub Actions <actions@github.com>
* option to source custom code in CLI * removed empty dicts * fixed issue with removed tensor metadata in match case statements
* forbid extra fields in config * fixed configs
* automatic inference of attach index based on type signature * added inference for input and x names
* fixed link in docs * fixed repvgg backbone * fixed efficientnet
* upload logs to mlflow * added mlflwo instance * multithread log upload * fixed upload logs * fixed log file path * removed exceptions * logging exceptions * fixed typo * reverted exception * moved line * replaced warning with error log * Update trainer.py
* add archiver CLI * add archiver callback * add max_det parameter to EfficientBBoxHead * add enum to categorize tasks for the implemented heads * add archiver tests * adjust Archiver to new nn archive format * pre-comit formatting * add LDF creation and adjust to new nn archive format * update requirements.txt * add opencv-python to requirements.txt * add support for ImplicitKeypointBBoxHead * remove support for ObjectDetectionSSD * Update requirements.txt * Added mlflow and removed opencv * [Automated] Updated coverage badge * add support for SegmentationHead and BiSeNetHead * base archiver tests on model from luxonis-train instead of torchvision * adjust head parameters to changes in NN Archive * adjust keypoint detection head parameters to changes in NN Archive * bugfix - make sure self.max_det is used in nms * add max_det parameter to ImplicitKeypointBBoxHead * adjust task categorization for ImplicitKeypointBBoxHead * fixing Windows PermissionError occuring on file deletion * fixing Windows PermissionError occuring on file deletion due to unreleased logging handlers * add method to remove file handlers keeping the log file open * add a logging statement at the end of archiving * add optuna_integration to requirements.txt * add hard-coded solution to determining is_softmax parameter * added help --------- Co-authored-by: Martin Kozlovský <martin.kozlovsky@luxonis.com> Co-authored-by: GitHub Actions <actions@github.com>
* extend NN Archive generation test coverage to cover all implemented heads * [Automated] Updated coverage badge --------- Co-authored-by: GitHub Actions <actions@github.com>
* uploading all checkpoints * fix names * removed comment
* handling SIGTERM signal * resume argument takes path
* handling SIGTERM signal * resume argument takes path * basic task group labels support * updated requirements * fixed tests * fixed loader test * Update luxonis_train/models/luxonis_model.py Co-authored-by: conorsim <60359299+conorsim@users.noreply.github.com> --------- Co-authored-by: conorsim <60359299+conorsim@users.noreply.github.com>
* option to set torch matmul precision for tensor cores * updated readme
* fixed reset not being called * added metric resets * removed inheritance * proper oks reset * removed unnecessary resets * added annotations
* Added seed to config for reproducibility * fixed seg drawing when using torch deterministic backend * added deterministic order of creating nodes * removed seed from example config * added reproducability to inspect * formatting --------- Co-authored-by: DrejcPesjak <dp8949@student.uni-lj.si> Co-authored-by: Martin Kozlovský <martin.kozlovsky@luxonis.com>
* support for custom loaders and datasets * updated configs * custom loaders in inspect command * updated inspect for multi-task labels * removed custom loader from test config * deleted comment * deleted comment * removed custom dataset * removed comment * skipping archiver test untill fixed in luxonis-ml * [Automated] Updated coverage badge --------- Co-authored-by: GitHub Actions <actions@github.com>
Co-authored-by: GitHub Actions <actions@github.com>
Co-authored-by: GitHub Actions <actions@github.com>
Co-authored-by: GitHub Actions <actions@github.com>
Co-authored-by: GitHub Actions <actions@github.com> Co-authored-by: Martin Kozlovsky <martin.kozlovsky@luxonis.com>
Co-authored-by: Martin Kozlovsky <martin.kozlovsky@luxonis.com> Co-authored-by: Michal Sejak <hyas@seznam.cz> Co-authored-by: GitHub Actions <actions@github.com>
Co-authored-by: Martin Kozlovsky <martin.kozlovsky@luxonis.com>
Test Results2 files 2 suites 18s ⏱️ For more details on these errors, see this check. Results for commit 4265845. |
Co-authored-by: GitHub Actions <actions@github.com>
kozlov721
requested review from
tersekmatija
and removed request for
a team
kozlov721
requested changes
Sep 20, 2024
| @@ -257,3 +267,135 @@ def _get_final_assignments( | |||
| ) | |||
|
|
|||
| return assigned_labels, assigned_bboxes, assigned_scores | |||
|
|
|||
|
|
|||
| class RotatedTaskAlignedAssigner(TaskAlignedAssigner): | |||
There was a problem hiding this comment.
I'd move this to a separate file
There was a problem hiding this comment.
And the name could be changed to OrientedTaskAlignedAssigner to be consistent with the OBB naming.
| return out_target | ||
|
|
||
|
|
||
| class VarifocalLoss(nn.Module): |
There was a problem hiding this comment.
The VarifocalLoss is already implemented in luxonis_train.attached_modules.losses.adaptive_detection_loss
luxonis_train/assigners/utils.py
Outdated
| """Select the positive anchor center in gt for rotated bounding | ||
| boxes. | ||
|
|
||
| Args: |
There was a problem hiding this comment.
The docstring must be in epytext
| """ | ||
| super().__init__(**kwargs) | ||
|
|
||
| if not isinstance(self.node, EfficientOBBoxHead): |
There was a problem hiding this comment.
This is not necessary; the check is done automatically in BaseAttachedModule.
| gt_cxcywhr = target[:, :, 1:] | ||
| mask_gt = (gt_cxcywhr.sum(-1, keepdim=True) > 0).float() | ||
|
|
||
| # TODO: log change of assigner (once common Logger) |
There was a problem hiding this comment.
There is no change of assigner to log
| return stack([pt1, pt2, pt3, pt4], -2) | ||
|
|
||
|
|
||
| def xyxy2xywh(x: Tensor) -> Tensor: |
There was a problem hiding this comment.
There's already torchvision.ops.box_convert for this
| return y | ||
|
|
||
|
|
||
| def xywh2xyxy(x: Tensor) -> Tensor: |
There was a problem hiding this comment.
torchision.ops.box_convert
| @@ -269,6 +405,128 @@ def bbox_iou( | |||
| return iou | |||
|
|
|||
|
|
|||
| def probiou( | |||
There was a problem hiding this comment.
more readable function name
| return iou | ||
|
|
||
|
|
||
| def batch_probiou(obb1: Tensor, obb2: Tensor, eps: float = 1e-7) -> Tensor: |
There was a problem hiding this comment.
This function is almost identical to probiou, why do we need both?
| @@ -411,6 +669,182 @@ def non_max_suppression( | |||
| return output | |||
|
|
|||
|
|
|||
| def non_max_suppression_obb( | |||
There was a problem hiding this comment.
This is almost identical to non_max_suppression. Can you improve non_max_suppression to also support oriented bboxes instead of having this function?
Co-authored-by: Martin Kozlovsky <martin.kozlovsky@luxonis.com>
Co-authored-by: Martin Kozlovsky <martin.kozlovsky@luxonis.com> Co-authored-by: GitHub Actions <actions@github.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.
This PR introduces oriented bounding box functionality for object detection tasks, built on top of
DetectionModel.Implemented classes