Authors: Nicolas Harvey Chapman, Feras Dayoub, Will Browne, Chris Lehnert

Published on: February 06, 2024

Impact Score: 8.15

Arxiv code: Arxiv:2402.03721

Summary

• What is new: A novel implicit object memory method for embodied object detection, leveraging projective geometry to improve detection in robots.
• Why this is important: Existing single-image object detectors do not perform optimally in robotics due to lacking context from complex multi-modal data streams.
• What the research proposes: An image object detector pre-trained on language-image data, extended with an implicit object memory to use spatial and temporal data for better detection.
• Results: Improvement of 3.09 mAP over existing VOD and Semantic Mapping techniques, and a significant 16.90 mAP increase over baselines not pre-trained on language-image data.

Technical Details

Technological frameworks used: Implicit object memory for embodied object detection

Models used: Pre-trained language-image data models

Data used: Embodied data streams from diverse indoor scenes

Potential Impact

Robotics, smart home devices, surveillance, autonomous vehicles

Want to implement this idea in a business?

We have generated a startup concept here: SightWise.