tensorflow / tensorflow/datasets
object detection datasets pads images for batch, can't unnormalize bbox without original image shape
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 4.6k
- Forks
- 1.6k
- Avg merge
- 3h 54m
- Merged PRs (30d)
- 1
Description
Object detection datasets like voc,coco provide images and normalized bboxes(0-1). When applied with batch, it produces batch with largest image shape and pads all other images to that shape. BBoxes are not rescaled accordingly. It also doesn't provide original image shape to unnormalize bbox to image shape leading to wrong bboxes. Is there any other way to avoid this?
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
The issue does not name a source file, test, or entry point. Start by tracing object-detection dataset batching, image padding, and bounding-box normalization, then identify how the original image shape is represented; done means batched boxes can be mapped back to each image without incorrect coordinates and the behavior is covered by a regression test.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, tensorflow
- Domain
- data, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100