See, Act, Adapt: Active Perception for Unsupervised Cross-Domain Visual Adaptation via Personalized VLM-Guided Agent

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• Sea² improves visual grounding performance by 13.54% on the ReplicaCAD dataset compared to baseline methods.result 90%
• Sea² improves segmentation performance by 15.92% on the ReplicaCAD dataset compared to baseline methods.result 90%
• Sea² improves 3D box estimation performance by 27.68% on the ReplicaCAD dataset compared to baseline methods.result 90%
• Sea² keeps all perception modules frozen during adaptation, requiring no downstream labels during training.method 85%
• Sea² transforms a vision-language model into a low-level pose controller through a two-stage training pipeline.method 80%
• Sea² uses only scalar perceptual feedback to navigate the agent toward informative viewpoints.technical 80%
• Sea²'s performance depends on the quality of pre-trained models, and significant capability gaps may hinder viewpoint optimization.limitation 75%
• Sea² directly leverages off-the-shelf perception models for various tasks without the need for retraining.technical 80%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card