The authors adopt the Learning Using Privileged Information (LUPI) paradigm.
Privileged motion and gravity cues improve feedforward 3D reconstruction without extra test-time cost.
Training with residual optical flow and gravity direction as privileged signals makes transformer-based reconstruction more robust in dynamic scenes, with no change to inference inputs.
Chinese Tech
Shiyao Qian · Yuan Ren · Dongfeng Bai
Huawei Noah's Ark Lab · University of Toronto
Research Digest··2 min read
Thread:3D Object Reconstruction
Qian, Ren and Bai augment feedforward transformer models for multi-view 3D reconstruction by supplying optical flow and gravity direction as privileged information during training only.
Why this paper
From Huawei Noah's Ark Lab and University of Toronto · Part of 3D Object Reconstruction, now 9 papers
In one line
Adding training-only optical flow and gravity cues makes feed-forward 3D reconstruction more accurate and robust without extra computation at inference.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§