Tetromino Pixel
"Tetromino Pixel," a term encompassing various research directions, broadly focuses on leveraging pixel-level information from images and videos to achieve higher-level tasks. Current research emphasizes using deep learning models, including transformers, U-Nets, and diffusion models, to process visual data and integrate it with other modalities like text and 3D point clouds for applications such as image captioning, object detection, 3D reconstruction, and robotic control. This work is significant for advancing multimodal AI, improving the efficiency and interpretability of computer vision systems, and enabling new capabilities in areas like autonomous navigation and medical image analysis.
Papers
Beyond Pixels: Text Enhances Generalization in Real-World Image Restoration
Haoze Sun, Wenbo Li, Jiayue Liu, Kaiwen Zhou, Yongqiang Chen, Yong Guo, Yanwei Li, Renjing Pei, Long Peng, Yujiu Yang
Explaining Object Detectors via Collective Contribution of Pixels
Toshinori Yamauchi, Hiroshi Kera, Kazuhiko Kawamoto