Sync to 3.3 #11602
Closed
Sync to 3.3 #11602
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
This data structure is close to the DetDataSample. It removes the fields unrelated to one stage detection. Plus we added a reid_labels field instances.
Add loading and transforms for ReID Detection annotations ReIDDataSample - Add LoadReIDDetAnnotations loading, with its test. This is a child transform of LoadAnnotations. It changes the _load_labels method to include the person ids for each detections - Add PackReIDDetInputs formatting, with its test. Since the core SampleData is not a DetDataSample, we re-implement the transform and refactor it. This format should be compatible with Single Stage detector.
Other pack transforms use img_id for the ID of an image which is not compliant with the annotations of openmmlab 2.0. So we renamed this meta key to img_label.
There were several incompatibilties from annotations and the loading/formatting steps. - Loading transforms has been adapted to match cuhk SYSU PEDES annotations source. - Formatting has been arranged, the output stays compatible with mmdet detectors.
The triplet loss is 0 if there is only one positive anchor and nothing else in the the instances.
The dataset only works with open mmlab annotations format. Only train loader is implemented in the configs.
It does not support the parameter 'with_bbox_refine' from Deformable DETR.
Add of the PSTR model and its base class (Detection ReID). Also, update of the abstract structure for instance reid detection. We made the decision to have it implicit so we can keep the original `InstanceData` as the class. We might use it in the evaluation function. As it is currently, it only to provide clear information about the new instance data structure.
not working
- Dockerfile has g++ - relative paths for base config - remove useless entrypoint from docker-compose
Edit Dockerfile and compose to permit a clean two stages image for training. The first stage install mmcv (and mmengine) with CUDA support. The image is 14+GB. The second image is taking the compiled version from the first stage and install the mmdet repo on a specific checkout. It "only" takes 8+GB, thanks to the torch image is running on its runtime version. The size of the images are only illustrative. It heavily depends of the torch base images' sizes.
- adapt the num_persons from the reid head to match the new train set. - refactor the pstr reid head sample/batch: - the weight is set in the sample loss function - the management of the dict for one sample is clearer
oim sum the loss by detections
refactoring the runtime in two step. One for the dependencies and another one to get mmdet from build stage.
Train docker build
|
AwePhD seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account. You have signed the CLA already but the status is still pending? Let us recheck it. |
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.
No description provided.