Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Videos

Competing for Pixels: A Self-Play Algorithm for Weakly-Supervised Semantic Segmentation.

Shaheer U Saeed, Shiqi Huang, Joao Ramalhinho

IEEE Transactions on Pattern Analysis and Machine Intelligence

|October 3, 2024

Summary

This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Associations of indoor bacterial, metabolomic, and other chemical exposures with allergic rhinitis in school-aged children.

BMC microbiology·2026

Same author

Reasoning in machine vision by learning fast and slow thinking.

Nature communications·2026

Same author

A bilayer theranostic hydrogel integrating visual pH monitoring with synergistic diabetic wound healing treatment.

Journal of materials chemistry. B·2026

Same author

Worldwide prevalence of diabetic ketoacidosis at diagnosis of type 1 diabetes: A systematic review and meta-analysis.

Preventive medicine·2026

Same author

Depth-based local registration refinement for augmented reality in pituitary surgery.

International journal of computer assisted radiology and surgery·2026

Same author

Harnessing cellular immunotherapy for cholangiocarcinoma: an integrated roadmap for overcoming resistance.

Frontiers in immunology·2026

Same journal

HardFlow: Hard-Constrained Sampling for Flow-Matching Models Via Trajectory Optimization.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

Industrial Brain: Self-Evolving Neuro-Symbolic Autonomy with Causal Resilience for Cyber-Physical Systems.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

Adaptive Hardness-Driven Dictionary Distillation for Incomplete Streaming View Clustering.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

Mixture of Global and Local Experts with Diffusion Transformer for Controllable Face Generation.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

Task-KV: Task-aware KV Cache Optimization via Semantic Differentiation of Attention Heads.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

Achieving Text-based Person Retrieval with Any Granularity.

IEEE transactions on pattern analysis and machine intelligence·2026

See all related articles

This study introduces a novel weakly-supervised semantic segmentation (WSSS) method using reinforcement learning (RL) self-play to gamify image segmentation. The approach trains agents to compete for object patches, significantly improving segmentation accuracy and reducing common WSSS errors.

Area of Science:

Computer Vision
Artificial Intelligence
Machine Learning

Background:

Weakly-supervised semantic segmentation (WSSS) relies on image-level labels, lacking precise region correspondence.
WSSS offers reduced annotation costs compared to fully-supervised methods.
Existing WSSS methods struggle with over-segmentation and under-segmentation.

Purpose of the Study:

To propose a novel WSSS method using reinforcement learning (RL) self-play.
To gamify the image segmentation process for regions of interest (ROIs).
To address the challenge of accurate ROI localization with limited supervision.

Main Methods:

Formulated segmentation as a competitive game between two RL agents.
Agents select ROI-containing patches until all such patches are exhausted.

Related Experiment Videos

Utilized an object presence detector pre-trained on image-level binary labels.

Implemented a game termination condition and agent incentives based on patch selection success.

Main Results:

Demonstrated significant performance improvements across four diverse datasets.
Outperformed recent state-of-the-art WSSS methods.
Effectively minimized over- and under-segmentation issues inherent in WSSS.

Conclusions:

The proposed RL self-play gamified segmentation approach is effective for WSSS.
This method offers a robust solution for accurate ROI segmentation with weak supervision.
The competitive agent framework successfully mitigates common WSSS segmentation errors.