Takayuki Okatani
Takayuki Okatani studies how visual AI can understand practical scenes, with research on infrastructure inspection, driving hazards and image description. Okatani coauthored Grit, an image-captioning Transformer that uses two kinds of visual features. The work connected image processing with language generation, a recurring theme in his later multimodal research. Okatani directs RIKEN’s Multimodal Visual Intelligence Team. The team combines visual information with other forms of data. Its stated applications include bridge and road inspection, autonomous driving and driver assistance.
- Now at
- RIKEN and Tohoku University
- Works
- 241
- Citations
- 4,206
- h-index
- 33
Publication numbers from OpenAlex, as of 14 October 2026.
Timeline
-
2022
Generating image descriptions
Okatani coauthored Grit, an image-captioning Transformer that uses two kinds of visual features. The work connected image processing with language generation, a recurring theme in his later multimodal research.
-
2023
Inspecting faults in images and signals
Shared papers studied texture anomaly detection, adaptation for image segmentation and accelerometer-based monitoring of fatigue cracks. These projects applied learning methods to finding unusual patterns and inspecting physical structures.
-
2024
Connecting visual questions with inspection
His publications include shared work on estimating bridge damage causes from several images through visual question answering, predicting driving hazards with multimodal AI, and a lightweight image-classification network for single-board computers.
-
As of October 2026
Multimodal visual intelligence
Okatani directs RIKEN’s Multimodal Visual Intelligence Team. The team combines visual information with other forms of data. Its applications include bridge and road inspection, autonomous driving and driver assistance.
Sources and credits
Photo: www.riken.jp; creator not stated. License: Not stated by source. Source: riken.jp. Cropped to a square and resized. Ask us to remove or replace this photo.