Panasonic HD's Two Papers Accepted at CVPR 2026, a Top AI and Computer Vision Conference

Panasonic Holdings Corporation (Panasonic HD) announced that two of its research papers have been accepted for CVPR 2026, the world's premier international conference in the field of AI and computer vision. One of these papers was selected as a 'Highlight' for its exceptional research quality. The findings will be presented at the conference, which will be held in Colorado, USA, from June 3 to June 7, 2026. [Overview of Accepted Papers] ■ Paper 1: Highly Efficient Spatial Recognition Technology Supporting the Practical Application of Physical AI By efficiently compressing 3D spatial information, this technology achieves both a reduction in processed data volume and high spatial recognition capability. It contributes to the advancement of AI operating in the real world, such as robotics and physical AI. In recent years, there has been growing attention toward 'Physical AI,' which enables robots and machines to recognize and judge real-world environments and act autonomously. Achieving this requires advanced spatial recognition capabilities, such as understanding the positional relationships between objects, and further evolution of multimodal AI is expected. However, conventional spatial recognition technologies using multimodal AI have faced the challenge of increased computational volume required to maintain spatial information. This technology combines highly efficient feature representation compression via clustering with step-by-step spatial recognition learning. It achieves spatial recognition performance equal to or better than other methods while suppressing the amount of spatial information handled by multimodal AI. For example, while some conventional 3D spatial recognition methods input approximately 8,000 tokens of spatial information into multimodal AI, this technology represents 3D space with 700 tokens. This technology contributes to practical applications in a wide range of fields, including future real-time processing for AI operating in the real world and areas requiring 3D spatial recognition and understanding of positional relationships. ■ Paper 2: Portable Active Learning (PAL) for Efficient Learning, Reducing AI Development Costs and Time This technology achieves high-precision object detection while significantly reducing annotation costs, which are the biggest bottleneck in AI development. At CVPR 2026, it was selected as a 'Highlight' among accepted papers, highly evaluated for its novelty, technical maturity, and future potential. AI-based image recognition technology is being utilized in various fields such as autonomous driving, factory inspection, and surveillance systems. However, developing high-performance AI requires 'annotation'—the manual, detailed labeling of 'what is where' in a large number of images—which poses challenges in terms of time and cost. 'Portable Active Learning (PAL)' developed this time integrates and evaluates multiple factors such as uncertainty, image diversity, and class imbalance to automatically determine 'which images the AI should prioritize learning.' As a result, it achieved equal or better recognition performance with approximately 20% less annotation work on average compared to conventional methods. Furthermore, this technology features a plug-and-play design that can be applied directly to various AI object detection models, eliminating the need for model modifications that were a challenge in conventional methods. It contributes to lower costs and increased efficiency in AI implementation in fields such as autonomous driving, edge AI, infrastructure inspection, and factory inspection. Panasonic HD will continue to accelerate the social implementation of AI and promote the research and development of AI technologies that contribute to the lives and workplaces of our customers.