Refactor heads to support multiple heads

[jin-s13]

fix #219

Migration note :

Config files need to be changed.

1. rename loss_pose to loss_keypoint.

2. move loss_keypoint from model settings to keypoint_head settings.

For example, in configs/top_down/hrnet/coco/hrnet_w32_coco_256x192.py.

from

keypoint_head=dict(
        type='TopDownSimpleHead',
        in_channels=32,
        out_channels=channel_cfg['num_output_channels'],
        num_deconv_layers=0,
        extra=dict(final_conv_kernel=1, ),
    ),
    train_cfg=dict(),
    test_cfg=dict(
        flip_test=True,
        post_process='default',
        shift_heatmap=True,
        modulate_kernel=11),
    loss_pose=dict(type='JointsMSELoss', use_target_weight=True))

to

keypoint_head=dict(
        type='TopDownSimpleHead',
        in_channels=32,
        out_channels=channel_cfg['num_output_channels'],
        num_deconv_layers=0,
        extra=dict(final_conv_kernel=1, ),
        loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
    train_cfg=dict(),
    test_cfg=dict(
        flip_test=True,
        post_process='default',
        shift_heatmap=True,
        modulate_kernel=11))

[codecov]

Codecov Report

Merging #382 (c45a387) into master (9f077d8) will decrease coverage by 0.48%.
The diff coverage is 77.65%.

@@            Coverage Diff             @@
##           master     #382      +/-   ##
==========================================
- Coverage   82.44%   81.96%   -0.49%     
==========================================
  Files         121      124       +3     
  Lines        8028     8235     +207     
  Branches     1298     1330      +32     
==========================================
+ Hits         6619     6750     +131     
- Misses       1145     1203      +58     
- Partials      264      282      +18     

Flag	Coverage Δ
unittests	81.96% <77.65%> (-0.49%)	⬇️

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

Impacted Files	Coverage Δ
...datasets/datasets/top_down/topdown_mpii_dataset.py	40.16% <0.00%> (-0.67%)	⬇️
...sets/datasets/top_down/topdown_mpii_trb_dataset.py	46.87% <0.00%> (-1.52%)	⬇️
...s/datasets/top_down/topdown_posetrack18_dataset.py	32.87% <0.00%> (-2.34%)	⬇️
...keypoint_heads/bottom_up_higher_resolution_head.py	83.76% <13.63%> (-16.24%)	⬇️
...pose/datasets/datasets/hand/interhand2d_dataset.py	77.08% <28.57%> (-4.94%)	⬇️
...ose/models/keypoint_heads/bottom_up_simple_head.py	72.91% <59.09%> (-4.12%)	⬇️
mmpose/apis/inference.py	65.26% <60.00%> (ø)
mmpose/models/detectors/top_down.py	65.28% <63.63%> (-14.16%)	⬇️
mmpose/datasets/datasets/hand/hand_base_dataset.py	93.63% <71.42%> (-1.51%)	⬇️
mmpose/models/detectors/bottom_up.py	62.40% <76.19%> (-1.57%)	⬇️
... and 22 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 9f077d8...c45a387. Read the comment docs.

[innerlee]

Give a clear migration note on the first post. e.g., #288

[innerlee]

For better migration experience, we can keep loss_pose here, and raise an error if it is not None, and give the link to this pr in the error message.

[innerlee]

Two comments.