pytorch / pytorch/vision

[feature request] ImageNet dataset to (optionally) return detections

Open
#5,318 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

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

Description

🚀 The feature

ImageNet dataset features extra annotated bounding box detections for some classes: https://image-net.org/download-bboxes.php. It would be nice if torchvision supported parsing this bboxes_annotations.tar.gz tarball and returning these detections (e.g. as a transform).

Currently it's hard to do because ImageNet / ImageFolder / DatasetFolder do not return original file name as part of targets, so enriching the dataset is impossible to do just in a form of transform (and subclassing / patching is needed).

Motivation, pitch

N/A

Alternatives

No response

Additional context

No response

cc @pmeier

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 by examining the ImageNet, ImageFolder, and DatasetFolder implementations and the referenced ImageNet bounding-box archive format. Determine how optional detection annotations can be associated with original filenames and exposed by the dataset or a transform; done means the feature works without requiring subclassing or patching and preserves existing behavior when detections are not requested.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
computer-vision, machine-learning
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.