Artificial intelligence vision field has recently made significant progress. SenseTime officially announced the open-source of its multi-task vision model SenseNova-Vision-7B-MoT, aiming to provide a strong foundation for visual understanding and GUI intelligent agents development.
The open-sourced model has excellent task integration capabilities. It successfully integrates various core vision tasks, such as object detection, OCR (optical character recognition), depth estimation, normal estimation, image segmentation, and multi-view processing, into a single 7B parameter model. This architecture not only enhances the model's versatility but also supports defining new vision task variants through natural language, achieving flexible reconfiguration of vision capabilities beyond traditional task boundaries.
To support community research and applications, SenseTime adopted a fully open-source strategy. Developers can not only obtain model weights but also access a subset of the SenseNova-Vision corpus containing 50 million samples, as well as a complete reproduction toolkit for processing the remaining public data.
Industry experts point out that the release of SenseNova-Vision-7B-MoT is not a simple lightweight model iteration, but rather provides a more scalable unified framework for visual understanding tasks. Currently, the technical details of this model, code repositories, and demonstration spaces are available on platforms such as Hugging Face and GitHub, allowing researchers and developers to directly call them, accelerating technological innovation and practical applications in related fields.
Project address: https://github.com/OpenSenseNova/SenseNova-Vision
