Click3R corrects stereo 3D reconstruction with sparse clicks
Original title: Click3R: Interactive Stereo 3D Reconstruction with Sparse Correspondence Clicks
A lightweight adapter uses user point correspondences, and the authors say one click can cut reconstruction error.
Click3R adds interactive geometric correction to feed-forward stereo 3D reconstruction models, which otherwise emit a single prediction that cannot be revised. Given an image pair and sparse cross-view correspondences, a lightweight point correspondence adapter injects those constraints into the network. The authors say this helps with repetitive textures, symmetry, large viewpoint changes, and textureless regions, and that error falls with as few as one click on standard and a new failure-mode benchmark.
- Users supply a small set of cross-view point correspondences.
- The method targets ambiguities that feed-forward models cannot revise.
- The authors curated a benchmark to expose existing reconstruction failures.
- They report gains on both standard benchmarks and their curated set.
From the original article
1 min read at huggingface.co
Despite impressive progress, existing feed-forward large 3D reconstruction models produce a single, irrevocable prediction. When these models fail, there is no mechanism to correct their mistakes. To address this limitation, we introduce Click3R, a framework for interactive geometric correction of feed-forward models using sparse human correspondence clicks.
By Hao Gu, Jin Yao, Zezhou Cheng · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.