Add a new example for video feature vector input

[xianyuanliu]

Description

This PR follows PR #291 and #292, upgrading DA Trainers for feature vector input. Trainers on video image input and video feature vector input have two main differences below:

1. Three modality choices in image input: RGB, Flow and joint (RGB+Flow), due to no Audio available now; Five in feature vector: RGB, Flow, Audio, joint and all (RGB+Flow+Audio).
2. One class type in image input: verb; Two in feature: verb, verb+noun.

The main changes are below:

1. Add Audio branch into the pipeline.
2. Update optimizer work with SGD and ADAM.
3. Add adaptive choice into source/target dataset sampling.
4. Add JSON logger for EPIC UDA challenge.
5. Reduce dummy network dimension in test_video_domain_adapter.

Status

Ready

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• In-line docstrings updated.
• Source for documentation at docs manually updated for new API.

[codecov-commenter]

Codecov Report

Merging #293 (2f7770f) into main (e42b4ec) will decrease coverage by 0.67%.
The diff coverage is 90.52%.

@@            Coverage Diff             @@
##             main     #293      +/-   ##
==========================================
- Coverage   92.98%   92.31%   -0.68%     
==========================================
  Files          45       45              
  Lines        4635     5163     +528     
==========================================
+ Hits         4310     4766     +456     
- Misses        325      397      +72     

Impacted Files	Coverage Δ
kale/pipeline/domain_adapter.py	94.08% <ø> (ø)
kale/loaddata/multi_domain.py	89.50% <53.84%> (-3.14%)	⬇️
kale/loaddata/video_multi_domain.py	72.12% <67.85%> (-2.69%)	⬇️
kale/loaddata/videos.py	88.53% <90.00%> (+1.16%)	⬆️
kale/pipeline/video_domain_adapter.py	91.70% <91.01%> (-5.24%)	⬇️
kale/embed/video_feature_extractor.py	95.34% <93.10%> (-1.27%)	⬇️
kale/loaddata/video_access.py	96.55% <94.44%> (-1.37%)	⬇️
kale/embed/attention_cnn.py	98.03% <98.38%> (+0.53%)	⬆️
kale/embed/video_selayer.py	97.32% <100.00%> (+0.32%)	⬆️
kale/loaddata/dataset_access.py	83.33% <100.00%> (ø)
... 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 e42b4ec...2f7770f. Read the comment docs.

[github-actions]

This pull request has been automatically marked as stale due to lack of activity.