arXiv:2611.25061v1 Announce Type: new Abstract: Humans can understand the 3D structure of scenes even from highly unfamiliar or extreme viewpoints. Cognitive science suggests this generalization stems from a top-down generative inference strategy. In contrast, state-of-the-art 3D reconstruction neural networks typically rely on amortized, feedforward inference. While fast, this pattern-matching approach degrades when faced with out-of-distribution (OOD) scene conditions unseen during training. Inspired by the generative nature of human vision, we propose Infer3D, an inverse generative modeling framework that recovers the latent variables of a generative model to best match the monocular input. To prevent degenerate flat outputs, we couple the generative image prior with a frozen 3D lifting network that constrains the optimization space to a valid, multi-view-consistent 3D manifold. Across a variety of controlled synthetic and real-world tasks, Infer3D demonstrates superior OOD reconstruction accuracy compared to feedforward networks trained on equal or larger datasets. Furthermore, mirroring how human perception requires more time to process confusing stimuli, we introduce an adaptive inference strategy that expends more test-time compute on ambiguous OOD examples while efficiently routing in-distribution inputs. Project website: https://amanikiruga.github.io/infer3d-site/
Infer3D: Inverse Generative Modeling Enables Robust 3D Reconstruction
A new arXiv paper (2611.25061v1) introduces Infer3D, an inverse generative modeling framework that recovers the latent variables of a generative model to match a monocular input, improving 3D reconstruction on out-of-distribution scenes over feedforward networks trained on equal or larger datasets. Infer3D couples a generative image prior with a frozen 3D lifting network to constrain optimization to a valid, multi-view-consistent 3D manifold, avoiding degenerate flat outputs, and adds an adaptive inference strategy that spends more test-time compute on ambiguous OOD examples while routing in-distribution inputs efficiently. The authors say the approach is inspired by top-down generative inference in human vision, which generalizes to unfamiliar viewpoints where amortized feedforward networks degrade.
Infer3D replaces amortized feedforward 3D reconstruction with inverse generative inference constrained by a frozen 3D lifting network, reporting higher out-of-distribution accuracy than feedforward baselines trained on equal or larger datasets, plus adaptive test-time compute routed only to ambiguou
Run your AI side-project on zahid.host
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.