Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

feat: Added torchvision photometric augmentations in artefact detection training #764

Merged
merged 6 commits into from
Dec 29, 2021
Merged

feat: Added torchvision photometric augmentations in artefact detection training #764

merged 6 commits into from
Dec 29, 2021

Conversation

SiddhantBahuguna
Copy link
Contributor

@SiddhantBahuguna SiddhantBahuguna commented Dec 27, 2021 •
edited
Loading

This PR makes the following modifications to the existing repo:

  1. It adds torchvision photometric augmentations : ColorJitter & GaussianBlur for the training of artefact detection.
  2. Updates a checkpoint with improved recall & segmentation on test data.

Summary:

Before augmentation:
mean precision 0.57
mean seg 0.62
mean recall 0.82

After augmentation:
mean precision 0.51
mean seg 0.67
mean recall 0.89

Any feedback is welcome!

@SiddhantBahuguna
Copy link
Contributor Author

Awaiting for the new checkout and the potential improvement in the results on the real world data before merging this PR.

@SiddhantBahuguna SiddhantBahuguna self-assigned this Dec 27, 2021
@SiddhantBahuguna SiddhantBahuguna added ext: references Related to references folder type: enhancement Improvement topic: object detection Related to the task of object detection labels Dec 27, 2021
@SiddhantBahuguna SiddhantBahuguna added this to the 0.5.0 milestone Dec 27, 2021
fg-mindee
fg-mindee suggested changes Dec 27, 2021
View reviewed changes
Copy link
Contributor

@fg-mindee fg-mindee left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks a lot!
Could you run a full training with this PR before we merge it?

So that we can add the new checkpoint in the PR if it improves performances 👌

references/obj_detection/train_pytorch.py Outdated Show resolved Hide resolved
references/obj_detection/train_pytorch.py Outdated Show resolved Hide resolved
fg-mindee
fg-mindee approved these changes Dec 27, 2021
View reviewed changes
Copy link
Contributor

@fg-mindee fg-mindee left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks! Mind adding the performance changes in the PR please?

@SiddhantBahuguna
Copy link
Contributor Author

Thanks! Mind adding the performance changes in the PR please?

Sure, although I have added recall that matters the most to us. I will include other metrics as well.

@codecov
Copy link

codecov bot commented Dec 29, 2021

Codecov Report

Merging #764 (fc64cca) into main (1dc3374) will decrease coverage by 0.02%.
The diff coverage is n/a.

Impacted file tree graph

@@            Coverage Diff             @@
##             main     #764      +/-   ##
==========================================
- Coverage   96.02%   96.00%   -0.03%     
==========================================
  Files         129      129              
  Lines        4807     4807              
==========================================
- Hits         4616     4615       -1     
- Misses        191      192       +1
Flag Coverage Δ
unittests 96.00% <ø> (-0.03%) ⬇️

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

Impacted Files Coverage Δ
doctr/models/obj_detection/faster_rcnn/pytorch.py 100.00% <ø> (ø)
doctr/models/builder.py 96.07% <0.00%> (-0.99%) ⬇️
doctr/transforms/modules/base.py 94.44% <0.00%> (ø)

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 1dc3374...fc64cca. Read the comment docs.

@SiddhantBahuguna SiddhantBahuguna merged commit c72da96 into main Dec 29, 2021
@SiddhantBahuguna SiddhantBahuguna deleted the photomet_aug_artefact branch December 29, 2021 14:02
@fg-mindee fg-mindee added type: new feature New feature framework: pytorch Related to PyTorch backend and removed type: enhancement Improvement labels Dec 31, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@fg-mindee fg-mindee fg-mindee approved these changes

Assignees

@SiddhantBahuguna SiddhantBahuguna

Labels
ext: references Related to references folder framework: pytorch Related to PyTorch backend topic: object detection Related to the task of object detection type: new feature New feature
Projects
None yet
Milestone
0.5.0
Development

Successfully merging this pull request may close these issues.
2 participants
@SiddhantBahuguna @fg-mindee