Metrics
2 File views/ downloads
13 Record Views
Preprint
Agentic Reasoning for Robust Vision Systems via Increased Test-Time Compute
arXiv
09/19/2025
Abstract
Developing trustworthy intelligent vision systems for high-stakes domains, e.g., remote sensing and medical diagnosis, demands broad robustness without costly retraining. We propose Visual Reasoning Agent (VRA), a training-free, agentic reasoning framework that wraps off-the-shelf vision-language models and pure vision systems in a Think--Critique--Act loop. While VRA incurs significant additional test-time computation, it achieves up to 40\% absolute accuracy gains on challenging visual reasoning benchmarks. Future work will optimize query routing and early stopping to reduce inference overhead while preserving reliability in vision tasks.
Files and links (2)
Details
- Title
- Agentic Reasoning for Robust Vision Systems via Increased Test-Time Compute
- Resource Type
- Preprint
- Publisher
- arXiv
- Format
- link and pdf
- Copyright
- © the authors
- Identifiers
- 99381512426906600
- Academic Unit
- Intelligent Systems and Robotics; Hal Marcus College of Science and Engineering
- Language
- English