open-mmlab/mmrotate

How to properly add an additional scene classification head.

Open

#735 opened on Feb 18, 2023

View on GitHub
 (1 comment) (0 reactions) (0 assignees)Python (641 forks)github user discovery
community discussiongood first issue

Repository metrics

Stars
 (2,141 stars)
PR merge metrics
 (No merged PRs in 30d)

Description

What's the feature?

Not a feature request but a burning question.

I'm new to the library and struggling to figure out how to do something quite simple: add a scene classification head. That is, beyond the object classes and their bounding boxes, I'd like to classify the scene of the image (a categorical variable with 8 values).

I've amended the dataset to load "scene_label" for a batch. I've subclassed an existing head and it now outputs three values instead of two: cls_score, bbox_pred, scene_probs.

However, It isn't clear to me what to change in order to handle the new scene_probs, the scene_label and compute the scene loss. What other functions or classes need to be created or changed for all this to work?

None of the tutorials/docs make this clear. At least for me. ChatGPT was also no help.

Any advice would be appreciated.

Any other context?

No response

Contributor guide

Research direction
Study the existing head implementations (e.g., RotatedRetinaHead) to understand the forward and loss methods. Modify the forward to output scene logits, then add a scene classification loss in the loss method. Update the dataset pipeline and config to include scene labels.
Tech stack
pythonpytorch
Domain
machine learningai
Issue type
Research
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Mostly clear
Prerequisites
PythonPyTorchmmrotate
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
50