improve Coco(Captions|Detection) implementation

[pmeier]

CocoCaptions is a special case of CocoDetection. While CocoDetection returns the complete annotation dictionary, CocoCaptions extracts the captions out of them. Thus, both implementations are functionally identical except for

vision/torchvision/datasets/coco.py

Line 74 in 22c548b

target = [ann['caption'] for ann in anns]

To make this easier to maintain, this lets CocoCaptions inherit from CocoDetection and only overwrites the annotation loading.

Blocked by #3416 to ensure the same functionality.

[codecov]

Codecov Report

Merging #3417 (1190b61) into master (2f4efd3) will increase coverage by 0.10%.
The diff coverage is 41.17%.

@@            Coverage Diff             @@
##           master    #3417      +/-   ##
==========================================
+ Coverage   75.13%   75.24%   +0.10%     
==========================================
  Files         105      105              
  Lines        9722     9708      -14     
  Branches     1563     1562       -1     
==========================================
  Hits         7305     7305              
+ Misses       1930     1916      -14     
  Partials      487      487              

Impacted Files	Coverage Δ
torchvision/datasets/coco.py	46.42% <41.17%> (+15.47%)	⬆️

---

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 2f4efd3...d7b059b. Read the comment docs.

[fmassa]

Thanks for improving this!