Faster RCNN Model + Pascal VOC DataModule

[teddykoker]

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

What does this PR do?

1. Implements the Pascal VOC Detection dataset as a DataModule
2. Implements Faster RCNN as a LightningModule

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[pep8speaks]

Hello @teddykoker! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2020-08-22 19:41:49 UTC

[codecov]

Codecov Report

Merging #157 into master will increase coverage by 1.10%.
The diff coverage is 86.07%.

@@            Coverage Diff             @@
##           master     #157      +/-   ##
==========================================
+ Coverage   90.72%   91.82%   +1.10%     
==========================================
  Files          86       86              
  Lines        4052     4049       -3     
==========================================
+ Hits         3676     3718      +42     
+ Misses        376      331      -45     

Flag	Coverage Δ
#unittests	91.82% <86.07%> (+1.10%)	⬆️

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

Impacted Files	Coverage Δ
pl_bolts/models/detection/faster_rcnn.py	79.62% <79.62%> (ø)
pl_bolts/datamodules/__init__.py	100.00% <100.00%> (ø)
pl_bolts/datamodules/dummy_dataset.py	100.00% <100.00%> (ø)
pl_bolts/models/detection/__init__.py	100.00% <100.00%> (ø)
pl_bolts/callbacks/self_supervised.py	72.72% <0.00%> (ø)
pl_bolts/models/self_supervised/__init__.py	100.00% <0.00%> (ø)
.../models/autoencoders/basic_vae/basic_vae_module.py	92.50% <0.00%> (ø)
..._bolts/models/self_supervised/cpc/cpc_finetuner.py
pl_bolts/models/self_supervised/ssl_finetuner.py
... and 2 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 efa1db0...e2b25cf. Read the comment docs.

[teddykoker]

@Borda didn't add test for cli, but added test training with DummyDetectionDataset, which I just added.

[williamFalcon]

We have a dummy dataset already. why is this one different?

[teddykoker]

Instead of a tensor label it returns a dict with values {"boxes": FloatTensor[N, 4], "labels": Int64Tensor[N]} as per the standard in pytorch for detection models. Should this be a choice in the existing dummy dataset?

[teddykoker]

Also VOCDetectionDataModule won't work right now as they modified an image in the dataset that effects the md5sum of the download. I added a fix to torchvision thats been approved: pytorch/vision#2589

[williamFalcon]

@teddykoker mind fixing the failing tests here?

[teddykoker]

@williamFalcon it looks like the minimal version tests are failing because I am using a feature of torchvision introduced in 0.7.0 (our requirements state torchvision>=0.5).

Our options are:

1. Keep torchvision>=0.5 and remove the ability to specify number of trainable backbone layers in the FasterRCNN
2. Upgrade torchvision >= 0.7 and keep code as is.

Let me know what you think.

[williamFalcon]

for tests we can make torchvision 0.7 but we need to get rid of the deps in the requirements.txt

[teddykoker]

test_detection is passing now; it looks like there are issues with BYOL and SimCLR.