Fixes multi-module feature extraction bugs and handling of Pytorch dataloaders, which return image-target pairs.

[lciernik]

This PR fixes the following bugs in the feature extraction pipeline extract_features

• BaseExtractor.extract_features:
  • When storing intermediate results previously, we reset features=defaultdict(list). This, unfortunately, deletes the intermediate results of other modules. Therefore changed to features[module_name]=[]
  • Paths need to be created before trying to store data: if module/feature does not exist, then an error is thrown.
  • We store intermediate file names. When save_in_one_file=True, output_dir is prepended again. Which incorrect.
• PyTorchExtractor calls super().extract_features --> need to add file_name_suffix and save_in_one_file as arguments to be able to pass it to the parent.

NEW: Dataloader modifier

PR adds a context manager for PyTorch feature extraction. Assumes that a dataloader either returns an image or a tuple of (image, *args) or a list of (image, *args). If the dataloader returns a tuple of (image, *args) or a list of (image, *args), it will return only the images. The class does not modify otherwise (e.g., specialized dataloaders).

[lucaeyring]

Looks good to me, thanks for the fixes!!

[MarcoMorik]

Looks good to me

[codecov]

Codecov Report

❌ Patch coverage is 54.00000% with 23 lines in your changes missing coverage. Please review.
✅ Project coverage is 75.36%. Comparing base (a1a1092) to head (bae13a4).
⚠️ Report is 5 commits behind head on master.

Files with missing lines	Patch %	Lines
thingsvision/core/extraction/base.py	8.00%	23 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #191   +/-   ##
=======================================
  Coverage   75.36%   75.36%           
=======================================
  Files          40       40           
  Lines        2139     2172   +33     
  Branches      272      276    +4     
=======================================
+ Hits         1612     1637   +25     
- Misses        429      437    +8     
  Partials       98       98           

Flag	Coverage Δ
unittests	75.36% <54.00%> (+<0.01%)	⬆️

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

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[MarcoMorik]

Great Tests

[LukasMut]

LGTM