refactor onnx export of two stage

[jshilong]

Motivation

The recent ONNX-related development is too quickly, which made the code hard to read, and there is a problem that the return type of the same function such as _get_bboxes in the DenseHead and BboxHead would be different when export to ONNX , which will greatly affect the future development of other instance task and #4912.

Modification

This PR moves all ONNX related code of two-stage model to a new function onnx_export in corresponding class.

BC-breaking (Optional)

None

[codecov]

Codecov Report

Merging #5205 (489f5d2) into master (ade08ba) will decrease coverage by 0.04%.
The diff coverage is 26.72%.

❗ Current head 489f5d2 differs from pull request most recent head a186d4c. Consider uploading reports for the commit a186d4c to get more accurate results

@@            Coverage Diff             @@
##           master    #5205      +/-   ##
==========================================
- Coverage   65.39%   65.35%   -0.05%     
==========================================
  Files         276      276              
  Lines       21229    21204      -25     
  Branches     3561     3522      -39     
==========================================
- Hits        13883    13857      -26     
- Misses       6572     6599      +27     
+ Partials      774      748      -26     

Flag	Coverage Δ
unittests	65.31% <26.72%> (-0.06%)	⬇️

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

Impacted Files	Coverage Δ
mmdet/core/export/pytorch2onnx.py	15.68% <ø> (ø)
mmdet/models/dense_heads/detr_head.py	88.99% <0.00%> (-0.42%)	⬇️
mmdet/models/roi_heads/cascade_roi_head.py	76.59% <0.00%> (ø)
...heads/roi_extractors/single_level_roi_extractor.py	65.38% <0.00%> (-4.01%)	⬇️
mmdet/models/roi_heads/standard_roi_head.py	50.31% <7.31%> (-12.09%)	⬇️
mmdet/models/dense_heads/rpn_head.py	56.25% <13.33%> (-17.79%)	⬇️
mmdet/models/detectors/single_stage.py	75.00% <14.28%> (-1.09%)	⬇️
mmdet/models/dense_heads/yolo_head.py	49.33% <18.18%> (-1.82%)	⬇️
mmdet/models/detectors/base.py	53.84% <20.00%> (-1.36%)	⬇️
mmdet/models/roi_heads/bbox_heads/bbox_head.py	76.59% <25.00%> (-3.23%)	⬇️
... and 8 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 ade08ba...a186d4c. Read the comment docs.

[RangiLyu]

Why is there a 13.7 MB tmp.onnx in this PR?