Feat/improve mot and bytetracker

[lauriebax]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

Be able to use ByteTracker with a pretrained detector.

Modification

• Make inference_mot more robust against configuration differences
• Don't cast images to float before running test_pipeline, mimicking detection pipelines
• Improve speed PackTrackInput
• Update ByteTracker to also return instance masks

BC-breaking (Optional)

Output of inference_mot might be different because we don't cast to float before resizing causing different interpolation to happen
Output of inference_mot might be different because more transforms than just Resize are now used.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMPreTrain.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[Copilot]

Pull request overview

This PR enhances ByteTracker to support instance masks and improves the robustness and performance of multi-object tracking (MOT) inference pipelines.

Key Changes:

• Extended ByteTracker to handle and return instance masks alongside bounding boxes
• Optimized PackTrackInput transformation with smarter memory layout handling (3-5x speedup)
• Made inference_mot more flexible by preserving more transforms from the test pipeline instead of only keeping Resize

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 9 comments.

File	Description
mmdet/models/trackers/byte_tracker.py	Added mask tracking capabilities throughout the tracking pipeline
mmdet/datasets/transforms/formatting.py	Optimized array operations by checking memory contiguity before transpose
mmdet/apis/inference.py	Removed float casting and improved transform filtering logic
tests/test_models/test_trackers/test_byte_tracker.py	Converted tests to pytest fixtures and added parameterized tests for mask support

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[illyakaynov-vbti]

this is a very complicated list comprehension. I looked at it for couple of minutes and I still not completely understand what it does. Especially the bit with any(x in transform['type']. Could you simplify it?

[lauriebax]

Not really, since a transform can also sometimes have a scope in front (mmdet.LoadImage vs. LoadImage).

[Jinior]

I'm okay with this comprehension

[Jinior]

I'm okay with this comprehension

[github-actions]

Package	Line Rate	Branch Rate	Complexity	Health
mmdet	100%	100%	0	✔
mmdet.apis	69%	49%	0	➖
mmdet.datasets	58%	39%	0	➖
mmdet.datasets.api_wrappers	33%	28%	0	❌
mmdet.datasets.samplers	54%	41%	0	➖
mmdet.datasets.transforms	74%	53%	0	➖
mmdet.engine	100%	100%	0	✔
mmdet.engine.hooks	63%	43%	0	➖
mmdet.engine.optimizers	98%	88%	0	✔
mmdet.engine.runner	96%	83%	0	✔
mmdet.engine.schedulers	100%	83%	0	✔
mmdet.evaluation	100%	100%	0	✔
mmdet.evaluation.evaluator	29%	0%	0	❌
mmdet.evaluation.functional	68%	50%	0	➖
mmdet.evaluation.metrics	57%	41%	0	➖
mmdet.models	100%	100%	0	✔
mmdet.models.backbones	80%	60%	0	✔
mmdet.models.data_preprocessors	84%	69%	0	✔
mmdet.models.dense_heads	82%	64%	0	✔
mmdet.models.dense_heads.clr_head	48%	43%	0	❌
mmdet.models.dense_heads.clr_head.losses	81%	50%	0	✔
mmdet.models.dense_heads.clr_head.utils	55%	50%	0	➖
mmdet.models.detectors	65%	40%	0	➖
mmdet.models.language_models	64%	33%	0	➖
mmdet.models.layers	87%	71%	0	✔
mmdet.models.layers.transformer	85%	60%	0	✔
mmdet.models.losses	85%	69%	0	✔
mmdet.models.mot	91%	59%	0	✔
mmdet.models.necks	67%	41%	0	➖
mmdet.models.reid	89%	62%	0	✔
mmdet.models.roi_heads	42%	22%	0	❌
mmdet.models.roi_heads.bbox_heads	66%	46%	0	➖
mmdet.models.roi_heads.mask_heads	86%	70%	0	✔
mmdet.models.roi_heads.roi_extractors	97%	91%	0	✔
mmdet.models.roi_heads.shared_heads	72%	33%	0	➖
mmdet.models.seg_heads	98%	92%	0	✔
mmdet.models.seg_heads.panoptic_fusion_heads	88%	55%	0	✔
mmdet.models.task_modules	73%	100%	0	➖
mmdet.models.task_modules.assigners	87%	66%	0	✔
mmdet.models.task_modules.coders	76%	51%	0	✔
mmdet.models.task_modules.prior_generators	79%	69%	0	✔
mmdet.models.task_modules.samplers	62%	30%	0	➖
mmdet.models.task_modules.tracking	90%	67%	0	✔
mmdet.models.test_time_augs	46%	21%	0	❌
mmdet.models.trackers	80%	58%	0	✔
mmdet.models.tracking_heads	97%	74%	0	✔
mmdet.models.utils	51%	36%	0	➖
mmdet.models.vis	94%	57%	0	✔
mmdet.structures	71%	34%	0	➖
mmdet.structures.bbox	89%	79%	0	✔
mmdet.structures.mask	69%	43%	0	➖
mmdet.testing	84%	68%	0	✔
mmdet.utils	40%	27%	0	❌
mmdet.visualization	79%	67%	0	✔
Summary	73% (33630 / 46296)	52% (6134 / 11800)	0	➖