facebookresearch / facebookresearch/detectron2

Bugs in Pascal Context GT visualization

Open
#5,392 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
34.7k
Forks
7.9k
PR merge metrics
No merged PRs in 30d

Description

## Instructions To Reproduce the 🐛 Bug:
1. Full runnable code or full changes you made:

details can refer to the failed pull requests: https://github.com/facebookresearch/detectron2/pull/5391

2. What exact command you run:
I run the tools/visualize_data and find the visualize.py in utils have an out of date numpy dtype.

4. __Full logs__ or other relevant observations:

![image](https://github.com/user-attachments/assets/cfdedf68-d43f-40d1-b7a1-26df60162f60)
this should be train(id:359) howver is overflow and refer to fly(id:359-256=103)

5. please simplify the steps as much as possible so they do not require additional resources to
run, such as a private dataset.

expect to handle classes more than 256, however Pascal Context 459 is overflow.

## Environment:

Python 3.10 numpy 1.22.4 torch 1.13.1

Contributor guide

Open the contributing guide

Research direction

Start by running tools/visualize_data and inspect the visualize.py utility mentioned in the report, using the Pascal Context example and the stated Python, NumPy, and PyTorch versions. Trace the dtype used for class IDs and verify that classes above 256, including train (id 359), are rendered with the correct label rather than overflowing to fly (id 103).

Written by the indexing model from the issue text.

Assessment

Tech stack
numpy, python
Domain
computer-vision
Issue type
Bug
Difficulty
2/5
Estimated time
1-3 hours
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.