發生了什麼事
Associate Professor Hiroyuki Kubo of the Graduate School of Informatics is developing new imaging hardware designed to capture data that standard cameras cannot perceive, aiming to improve AI performance. His research focuses on modifying the capture process—using projectors and cameras to visualize internal structures like blood vessels or fluid dynamics—rather than relying solely on software analysis of standard images. Additionally, Kubo is developing an interactive AI-human collaboration system for anime production that allows artists to select from AI-generated coloring suggestions.
Dr. Hiroyuki Kubo’s research shifts the focus of AI imaging from post-capture software analysis to the physical capture process itself. By manipulating lighting and camera positioning, he aims to extract data that reveals internal object properties, such as distinguishing between visually similar substances like milk and liquid soap based on how light scatters within them.
The research has demonstrated the ability to visualize blood vessels beneath the skin in real time without radiation by using a spatial offset between illumination and imaging. This method is being explored for potential applications in clinical and home healthcare settings, as well as for diagnosing vascular conditions like varicose veins.
In a separate project, Kubo is using cellulose nanofibers to visualize water flow patterns through polarization changes. This technique is intended to support infrastructure maintenance, such as bridge inspections, and the design of fuel-efficient ships.
Regarding anime production, Kubo has moved away from fully automated coloring after finding that even minor errors are unacceptable in professional workflows. He has developed an interactive system where AI proposes color candidates for human artists to review and select, aiming to balance efficiency with the high accuracy required in the industry.
為什麼這很重要
This research addresses a fundamental bottleneck in AI: the quality and depth of input data. By capturing internal properties of objects—such as light scattering in liquids or subsurface tissue structures—Kubo’s approach provides AI models with richer, more accurate datasets that standard surface-level photography cannot offer. This could lead to more reliable AI applications in medical diagnostics, infrastructure maintenance, and industrial design. Furthermore, his collaborative approach to anime production acknowledges the limitations of fully automated generative tools, proposing a model that prioritizes accuracy and efficiency in creative workflows.
Standard AI models often struggle with material identification because they rely on surface-level light reflection. By providing models with data that reflects internal structure, Kubo’s work could significantly improve the reliability of AI in fields where material composition is critical.
The shift toward hardware-software integration in imaging is relatively rare, as most researchers focus exclusively on software. Kubo’s background in physics allows him to bridge this gap, potentially creating new 'rules' for data acquisition that bypass the limitations of conventional photography.
The model for anime coloring represents a pragmatic approach to , focusing on augmenting human labor rather than replacing it. This model addresses the 'one-percent error' problem that often renders fully automated generative tools unsuitable for professional creative production.
互動機制：它實際上是如何運作的
以互動方式探索這項發展背後的基礎技術。
In AI, what are a model's "parameters"?
接下來看什麼
Kubo’s participation in Japan’s ' Accelerator Challenge: GENIAC' suggests ongoing development of his anime coloring tools. While the research is currently in the experimental and collaborative phase, future practical implications include the potential for compact, low-cost diagnostic tools for home healthcare and more efficient infrastructure monitoring. There is no documented timeline for commercial availability or specific product releases for these technologies.
Kubo is currently participating in the GENIAC project, a Japanese government-backed initiative to strengthen capabilities. This involvement is expected to accelerate the development of his research projects.
The research remains in the academic and collaborative development stage. There is no information regarding public access, pricing, or specific commercial deployment dates for the imaging hardware or the anime coloring software.
Future developments will likely focus on scaling these systems for practical use, such as integrating the imaging hardware into portable devices for home healthcare or disaster response.