[Refactor] Update DEKR training settings and results

[Ben-Louis]

trafficstars

Motivation

Modification

1. modify DEKR encoder
2. modify BottomupResize to make the image located at the center of the expanded image
3. update results for HRNet-w32 & HRNet-w48 on COCO & Crowdpose

Accuracy on COCO

Souce	Backbone	AP	AP50	AP75	AR	AR50
paper	HRNet-w32	68.0	86.7	74.5	73.0	-
MMPose-v0.29	HRNet-w32	68.0	86.8	74.5	72.8	89.7
MMPose-v1.0	HRNet-w32	68.6	86.8	75.0	73.5	89.8
paper	HRNet-w48	71.0	88.3	77.4	76.0	-
MMPose-v0.29	HRNet-w48	70.9	87.6	77.3	75.8	90.9
MMPose-v1.0	HRNet-w48	71.4	88.3	77.7	76.2	91.5

Accuracy on CrowdPose

Souce	Backbone	AP	AP50	AP75	AP(E)	AP(M)	AP(H)
paper	HRNet-w32	65.7	85.7	70.4	73.0	66.4	57.5
MMPose-v0.29	HRNet-w32	66.3	85.7	71.5	74.4	67.2	57.4
MMPose-v1.0	HRNet-w32	66.3	85.7	71.4	74.0	67.1	57.6
paper	HRNet-w48	67.3	86.4	72.2	74.6	68.1	58.7
MMPose-v0.29	HRNet-w48	68.2	86.9	73.6	75.6	69.0	60.0
MMPose-v1.0	HRNet-w48	67.8	86.5	73.0	75.8	68.4	58.7

BC-breaking (Optional)

Use cases (Optional)

Checklist

Before PR:

• [ ] I have read and followed the workflow indicated in the CONTRIBUTING.md to create this PR.
• [ ] Pre-commit or linting tools indicated in CONTRIBUTING.md are used to fix the potential lint issues.
• [ ] Bug fixes are covered by unit tests, the case that causes the bug should be added in the unit tests.
• [ ] New functionalities are covered by complete unit tests. If not, please add more unit tests to ensure correctness.
• [ ] The documentation has been modified accordingly, including docstring or example tutorials.

After PR:

• [ ] CLA has been signed and all committers have signed the CLA in this PR.

[codecov[bot]]

Codecov Report

Base: 80.35% // Head: 80.23% // Decreases project coverage by -0.13% :warning:

Coverage data is based on head (ce22f7d) compared to base (07715a5). Patch coverage: 84.41% of modified lines in pull request are covered.

:exclamation: Current head ce22f7d differs from pull request most recent head 818b701. Consider uploading reports for the commit 818b701 to get more accurate results

Additional details and impacted files

@@             Coverage Diff             @@
##           dev-1.x    #1901      +/-   ##
===========================================
- Coverage    80.35%   80.23%   -0.13%     
===========================================
  Files          209      209              
  Lines        12463    12464       +1     
  Branches      2105     2105              
===========================================
- Hits         10015    10000      -15     
- Misses        1996     2010      +14     
- Partials       452      454       +2     

Flag	Coverage Δ
unittests	80.23% <84.41%> (-0.13%)	:arrow_down:

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

Impacted Files	Coverage Δ
mmpose/models/heads/hybrid_heads/dekr_head.py	71.07% <0.00%> (ø)
mmpose/evaluation/metrics/coco_metric.py	95.95% <25.00%> (ø)
mmpose/codecs/utils/offset_heatmap.py	83.33% <55.55%> (-12.32%)	:arrow_down:
mmpose/datasets/datasets/body/mpii_dataset.py	65.57% <83.33%> (ø)
mmpose/datasets/datasets/body/mpii_trb_dataset.py	92.10% <84.61%> (ø)
mmpose/codecs/root_displacement.py	89.42% <100.00%> (+1.68%)	:arrow_up:
.../datasets/datasets/base/base_coco_style_dataset.py	89.28% <100.00%> (ø)
...asets/datasets/hand/coco_wholebody_hand_dataset.py	100.00% <100.00%> (ø)
mmpose/datasets/transforms/bottomup_transforms.py	79.88% <100.00%> (ø)
... and 8 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.

[ly015]

May need a rebase to fix the lint issue in CI.

[ly015]

We may need to rethink the name "RootDisplacement", which I am afraid is not straightforward for some users. What about SPR (Structured Pose Representation) as it's termed in the SPM paper?

[Ben-Louis]

We may need to rethink the name "RootDisplacement", which I am afraid is not straightforward for some users. What about SPR (Structured Pose Representation) as it's termed in the SPM paper?

I agree. It is better to keep its original name