[Feature] Sparsedemo

[kennymckormick]

Add sparse demo:

• By setting the new arg stride, you can get predictions for every stride * sample_size frames.
• A feature required by web demo since the original one is too slow.

[innerlee]

If I remembered correctly, the current demo implements an adaptive "stride". Conceptually there are two processes,

• one process reads the video and write the frames on the screen, regardless whether prediction is finished or not.
• one process fetch one current frame, and do prediction, and write results on screen, and fetch frame again, looping

@dreamerlin

[codecov]

Codecov Report

Merging #468 (13c691c) into master (4df447b) will decrease coverage by 0.00%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #468      +/-   ##
==========================================
- Coverage   84.69%   84.69%   -0.01%     
==========================================
  Files         118      118              
  Lines        8344     8357      +13     
  Branches     1366     1366              
==========================================
+ Hits         7067     7078      +11     
- Misses        930      932       +2     
  Partials      347      347              

Flag	Coverage Δ
unittests	84.68% <ø> (-0.01%)	⬇️

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

Impacted Files	Coverage Δ
...maction/models/roi_extractors/single_straight3d.py	84.90% <0.00%> (-0.47%)	⬇️
mmaction/core/evaluation/accuracy.py	93.18% <0.00%> (+0.03%)	⬆️

---

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 4df447b...13c691c. Read the comment docs.

[dreamerlin]

Is it the same with INPUT_STEP ?

[kennymckormick]

If I remembered correctly, the current demo implements an adaptive "stride". Conceptually there are two processes,

• one process reads the video and write the frames on the screen, regardless whether prediction is finished or not.
• one process fetch one current frame, and do prediction, and write results on screen, and fetch frame again, looping

@dreamerlin

Plz check my comments to @dreamerlin

[dreamerlin]

If I remembered correctly, the current demo implements an adaptive "stride". Conceptually there are two processes,

• one process reads the video and write the frames on the screen, regardless whether prediction is finished or not.
• one process fetch one current frame, and do prediction, and write results on screen, and fetch frame again, looping

@dreamerlin

It is for camera demo, not for this one.

[dreamerlin]

@innerlee cc.