SenseTime unveils U1 Pro and open-sources a 50M-sample vision dataset at WAIC 2026
机器之心 · wechat · 2026-07-21
At WAIC 2026, SenseTime reportedly put two releases at the center of its multimodal strategy: SenseNova U1 Pro and SenseNova-Vision.
SenseNova U1 Pro
- A native multimodal agent base model aimed at long-horizon delivery tasks.
- It uses a unified architecture where understanding, generation, and editing share the same representation space.
- The system supports near-lossless visual I/O, native Mixture-of-Transformer design, and separate training for text and pixels.
- SenseTime claims it can run dozens of agentic generation loops, keeping global style and local text synchronized.
SenseNova-Vision
- Reframes classic vision tasks such as detection, segmentation, depth estimation, and 3D reconstruction as native multimodal generation.
- Removes task-specific heads and uses one shared representation for text, pixels, semantics, and geometry.
- The company also open-sourced SN-VC-50M, a 50-million-sample visual supervision dataset.
- Reported results include SOTA structured visual understanding and strong performance versus specialist baselines on geometry and segmentation.
The article frames these releases as part of a broader shift from stitched-together vision pipelines toward a unified multimodal foundation model.
More from Multimodal
- 3D ResNet Paper Crosses 3,000 Citations Eight Years After CVPR 2018 — HirokatuKataoka · 2026-09-11
- Non-coder builds full-featured Android ComfyUI client with ChatGPT, submits to Google Play — ComfierUI · 2026-09-11
- FastH3-Live hits 22fps: acceleration node benchmarks and the --vram-headroom trick — spartong945 · 2026-09-11
- Midjourney style code share: --sref 2912175708 — tisch_eins · 2026-09-11
- Astra storyboards plus Minimax H3 per-shot generation boost video success rates — Hailuo_AI · 2026-09-11
- MiniMax H3 MAX nails cooking anime clips: 15-second curry demo with prompts shared — Hailuo_AI · 2026-09-11