(Speed up TopDown Inference) modified inference_top_down_model, make model able to run on batches of bounding box

[namirinz]

I have been modified _inference_single_pose_model to for-loop preprocess (center, scale) all bounding boxes (let say N bboxes) and stack them into batch_data. And then feed these batch_data on pose_model to get (N) keypoint at once.

And also modify inference_top_down_model to return output as close as the old one.

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

Merging #560 (c8a9f43) into master (8203f11) will decrease coverage by 0.87%.
The diff coverage is 70.76%.

❗ Current head c8a9f43 differs from pull request most recent head d69da36. Consider uploading reports for the commit d69da36 to get more accurate results

@@            Coverage Diff             @@
##           master     #560      +/-   ##
==========================================
- Coverage   82.25%   81.37%   -0.88%     
==========================================
  Files         154      163       +9     
  Lines       10436    11271     +835     
  Branches     1655     1810     +155     
==========================================
+ Hits         8584     9172     +588     
- Misses       1463     1662     +199     
- Partials      389      437      +48     

Flag	Coverage Δ
unittests	81.29% <70.76%> (-0.87%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...datasets/datasets/animal/animal_macaque_dataset.py	12.13% <12.13%> (ø)
mmpose/apis/inference.py	49.57% <35.18%> (-5.14%)	⬇️
mmpose/models/detectors/pose_lifter.py	74.62% <74.62%> (ø)
mmpose/datasets/pipelines/pose3d_transform.py	89.31% <77.77%> (-1.68%)	⬇️
...se/datasets/datasets/body3d/body3d_h36m_dataset.py	83.47% <78.94%> (-0.60%)	⬇️
.../models/keypoint_heads/temporal_regression_head.py	88.77% <79.62%> (-2.30%)	⬇️
mmpose/core/utils/regularizations.py	81.81% <81.81%> (ø)
...ose/datasets/datasets/animal/animal_fly_dataset.py	91.20% <91.20%> (ø)
.../datasets/datasets/animal/animal_locust_dataset.py	91.20% <91.20%> (ø)
...e/datasets/datasets/animal/animal_zebra_dataset.py	91.20% <91.20%> (ø)
... and 24 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8203f11...d69da36. Read the comment docs.

[innerlee]

Thanks!

The linting fails. Pls install pre-commit to do auto-formatting. Under the repo's root dir

pip install -U pre-commit
pre-commit install
pre-commit run --all-files

[namirinz]

Video comparison with single inference and batch inference
mmpose batch inference comparison

[jin-s13]

Please sign CLA (Contributor License Agreement). @namirinz

[namirinz]

Please sign CLA (Contributor License Agreement). @namirinz

Done.

[HoBeom]

#560
IndexError: tuple index out of range

[HoBeom]

python demo/top_down_pose_tracking_demo_with_mmdet.py \
    demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py \
http://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
    configs/top_down/resnet/coco/res50_coco_256x192.py \
    https://download.openmmlab.com/mmpose/top_down/resnet/res50_coco_256x192-ec54d7f3_20200709.pth \
    --video-path /dev/video0 \
    --show

Exception handling is not exist when the human bounding boxes are not detected

Use load_from_http loader
Use load_from_http loader
Traceback (most recent call last):
  File "demo/top_down_pose_tracking_demo_with_mmdet.py", line 177, in <module>
    main()
  File "demo/top_down_pose_tracking_demo_with_mmdet.py", line 132, in main
    pose_results, returned_outputs = inference_top_down_pose_model(
  File "/mmpose/mmpose/apis/inference.py", line 370, in inference_top_down_pose_model
    assert bboxes.shape[1] == 5
IndexError: tuple index out of range

[innerlee]

Thanks! Please open an issue so that it is easier to track progress

[HoBeom]

    if len(person_results) == 0:
        return pose_results, returned_outputs