Feed-Forward SceneDINO for Unsupervised Semantic Scene Completion

Stefan Roth, Christian Rupprecht, Daniel Cremers, T Munich, T Darmstadt, Oliver Hahn, Felix Wimbauer, Christoph Reich, Aleksandar Jevtić

arXiv · PDF · Open in the Atlas

Code that ran

We have not lifted any functions out of this paper's repositories yet, so there is nothing we have run. If it links a repository, it is listed below.

Abstract

4 MCML 5 ELIZA 6 hessian.AI * equal contribution https://visinf.github.io/scenedino Single Input Image SceneDINO 3D Feature Field Distill & Cluster SSC Prediction Figure 1. SceneDINO overview. Given a single input image (left), SceneDINO predicts both 3D scene geometry and 3D features in the form of a feature field (middle) in a feed-forward manner, capturing the structure and semantics of the scene. Unsupervised distillation and clustering of SceneDINO's feature space leads to unsupervised semantic scene completion predictions (right).

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2507.06230")
get_code_for_paper("2507.06230")
have("2507.06230")

Connect an agent — have() is free.