Pith. sign in

REVIEW 2 cited by

LEAP: Liberate Sparse-view 3D Modeling from Camera Poses

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 2310.01410 v1 pith:K4CWYYEL submitted 2023-10-02 cs.CV

LEAP: Liberate Sparse-view 3D Modeling from Camera Poses

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

Are camera poses necessary for multi-view 3D modeling? Existing approaches predominantly assume access to accurate camera poses. While this assumption might hold for dense views, accurately estimating camera poses for sparse views is often elusive. Our analysis reveals that noisy estimated poses lead to degraded performance for existing sparse-view 3D modeling methods. To address this issue, we present LEAP, a novel pose-free approach, therefore challenging the prevailing notion that camera poses are indispensable. LEAP discards pose-based operations and learns geometric knowledge from data. LEAP is equipped with a neural volume, which is shared across scenes and is parameterized to encode geometry and texture priors. For each incoming scene, we update the neural volume by aggregating 2D image features in a feature-similarity-driven manner. The updated neural volume is decoded into the radiance field, enabling novel view synthesis from any viewpoint. On both object-centric and scene-level datasets, we show that LEAP significantly outperforms prior methods when they employ predicted poses from state-of-the-art pose estimators. Notably, LEAP performs on par with prior approaches that use ground-truth poses while running $400\times$ faster than PixelNeRF. We show LEAP generalizes to novel object categories and scenes, and learns knowledge closely resembles epipolar geometry. Project page: https://hwjiang1510.github.io/LEAP/

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. Finding NeMO: A Geometry-Aware Representation of Template Views for Few-Shot Perception

    cs.CV 2026-02 conditional novelty 7.0

    A single network, given a few RGB views of an unseen object, builds a NeMO point-cloud representation that supports few-shot detection, segmentation, surface reconstruction, and 6DoF pose estimation without retraining.

  2. WorldMirror: Universal 3D World Reconstruction with Any-Prior Prompting

    cs.CV 2025-10 conditional novelty 5.0

    A unified feed-forward network that ingests optional geometric priors and jointly predicts point maps, depth, camera poses, normals, and 3D Gaussians, reporting state-of-the-art results on multiple 3D benchmarks.