Images rarely show an object’s full 3D geometry. At #ECCV2026, our research team introduced Axolotl3D, a multimodal and occlusion-aware 3D generation model. It combines images, camera data and partial geometry to reconstruct missing regions while preserving observed ones, achieving state-of-the-art results across single- and multi-view settings. Project page:

This can create a 3D

So you added depth to images or you generate models. We need a version that’s good at technical car

filling in the missing parts, cool tech

这种 3D 重建如果能稳定保留细小结构，对机器人和数字孪生都很实用。我会特别想看遮挡严重、视角很少时的失败案例。

Interesting to see the focus on inference efficiency here. The serving details matter as much as the model size.

yeah i like this. if youre filling in the parts you cant see, you gotta make sure the parts you can see stay accurate too

Reconstructing what a camera cannot see is a major challenge for 3D perception.Occlusion-aware models like this could strengthen digital twins, simulation, and robotics by creating more complete representations from limited visual data. 🤖📐

Strong fit for robotics and digital twins if the completed mesh stays watertight and physically plausible under simulation.

every 3D completion demo looks great on the axolotl and the dragon, ask me about the object nobody picked for the slide

This is a nice fit for asset cleanup. I’m curious how much camera metadata quality affects the reconstruction—does it fail gracefully with sparse input geometry, or just hallucinate detail?

📝

AI is slowly learnin that reality doesnt stop where the camera stops

这 3D 重建我很关心“没看到的部分”怎么处理只在单视图上看起来完整还不够最好能把遮挡区的置信度也出来；游做编或仿真时，知道哪里是猜的会差很。

handling partial geometry without blowing up inference latency is the real hurdle here

This is a major step for 3D AI. Reconstructing hidden geometry while preserving what’s actually observed could unlock powerful new workflows across vision, robotics, and simulation.