Pith. sign in

REVIEW 2 cited by

iFusion: Inverting Diffusion for Pose-Free Reconstruction from Sparse Views

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2312.17250 v1 pith:TGJMON2L submitted 2023-12-28 cs.CV

iFusion: Inverting Diffusion for Pose-Free Reconstruction from Sparse Views

classification cs.CV
keywords viewsdiffusionnovelreconstructionmodelobjectposeview
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We present iFusion, a novel 3D object reconstruction framework that requires only two views with unknown camera poses. While single-view reconstruction yields visually appealing results, it can deviate significantly from the actual object, especially on unseen sides. Additional views improve reconstruction fidelity but necessitate known camera poses. However, assuming the availability of pose may be unrealistic, and existing pose estimators fail in sparse view scenarios. To address this, we harness a pre-trained novel view synthesis diffusion model, which embeds implicit knowledge about the geometry and appearance of diverse objects. Our strategy unfolds in three steps: (1) We invert the diffusion model for camera pose estimation instead of synthesizing novel views. (2) The diffusion model is fine-tuned using provided views and estimated poses, turned into a novel view synthesizer tailored for the target object. (3) Leveraging registered views and the fine-tuned diffusion model, we reconstruct the 3D object. Experiments demonstrate strong performance in both pose estimation and novel view synthesis. Moreover, iFusion seamlessly integrates with various reconstruction methods and enhances them.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. ROAR-3D: Routing Arbitrary Views for High-Fidelity 3D Generation

    cs.CV 2026-05 unverdicted novelty 6.0

    ROAR-3D adds a token-wise view router and dual-stream attention to pretrained single-view 3D generators so they can use arbitrary unposed images for higher-fidelity output.

  2. Landscape-Awareness for Geometric View Diffusion Model

    cs.CV 2026-05 unverdicted novelty 4.0

    A score-based method is introduced to guide optimization in geometric view diffusion models toward correct viewpoints, improving convergence and sample efficiency over naive multistart strategies.