pytorch / pytorch/vision

Image augmentation with bounding boxes.

Open
#2,213 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

module: transforms topic: object detection
Dominant language
Python
Stars
17.9k
Forks
7.3k
Avg merge
1d 15h
Merged PRs (30d)
13

Description

🚀 Feature

Image augmentation with bounding boxes.

Motivation

Torchvision already has data augmentation for images, but I think it would be convinient to support images with bounding boxes as well.

Here is a jupyter notebook with an example: github.com/maximlopin/boxaug/blob/master/example.ipynb

Implementation

One way to implement this is to add an optional argument bboxes to __call__ methods, and some transforms will just ignore it (e.g. ColorJitter).

Alternatives

Wihout having this built-in, you have to use and learn other libraries. Torchvision already has a simple interface for image augmentation, and adapting bounding boxes support to the same interface would be very convinient.

cc @vfdev-5

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the linked boxaug example.ipynb and the existing torchvision image-transform interface. Determine which transforms must update bounding boxes and which can ignore them, then define the required API and coverage for the feature. Done means bounding-box-aware augmentation works through the same interface while preserving the expected image and box relationship.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
computer-vision, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.