SenseTime open-sources SenseNova-Vision unified vision model

SenseTime has released and fully open-sourced SenseNova-Vision, a unified vision model in its SenseNova foundation-model suite. The model is designed to handle object detection, optical character recognition, keypoint localization, image segmentation, depth estimation and 3D reconstruction within one system, instead of relying on separate specialist models for each task. The model also supports dense geometric prediction, including surface-normal estimation, as well as multi-view 3D tasks such as point-cloud reconstruction and camera-pose estimation. SenseTime has released the SenseNova-Vision Corpus-50M, a visual instruction dataset containing 50 million samples, and plans to integrate the model into its SenseNova U-series. [IT Home, in Chinese] Every Wednesday and Friday, TechNode’s Briefing newsletter delivers a roundup of the most important news in China tech, straight to your inbox. Your support helps TechNode continue to provide credible, on-the-ground journalism and industry insights about the Chinese tech industry.

Read the full article at the source →