XNews

SenseTime Open-Sources SenseNova-Vision: Native Unified Vision Model Outperforms Google Vision Banana

SenseTime Open-Sources SenseNova-Vision: Native Unified Vision Model Outperforms Google Vision Banana

SenseTime open-sourced SenseNova-Vision, a single large model that natively handles object detection, image segmentation, depth prediction, and 3D reconstruction. Unlike previous approaches that packaged separate expert models, it makes vision a native capability of the model. It matches or outperforms dedicated expert models and surpasses Google's Vision Banana in most benchmarks. The release includes model weights and a 50-million-sample visual instruction corpus. A single model covers high-frequency vision tasks, reducing deployment costs.

Source posts