Part 7/14:
Tencent’s release of Hunan World Mirror 1.1 demonstrates a significant leap in 3D reconstruction technology. The model can produce detailed 3D point clouds, depth maps, camera parameters, and surface normals from just a single image, video, or multi-view photos—all in real time on a single GPU.
This model is versatile: it can incorporate pre-calibrated camera intrinsics or lidar and RGBD sensor data to refine the output. By utilizing a transformer backbone and dedicated decoding heads, it efficiently processes spatial information, enabling applications in robotics, augmented reality, and scene understanding.