PulseAugur
LIVE 04:03:48
research · [2 sources] ·
0
research

PASR framework improves 3D shape retrieval using pose-aware analysis-by-synthesis

Researchers have introduced PASR, a novel framework for 3D shape retrieval from single, potentially occluded images. PASR distills knowledge from the DINOv3 foundation model into a 3D encoder, enabling alignment between pose-conditioned 3D projections and 2D feature maps. This approach formulates retrieval as an analysis-by-synthesis problem, where test-time optimization searches for the optimal shape and pose to reconstruct image features. The method demonstrates significant improvements over existing techniques in handling occlusions and offers multi-task capabilities including pose estimation and category classification. AI

Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →

IMPACT Enhances 3D shape retrieval robustness, potentially improving applications in AR/VR and robotics that rely on partial object recognition.

RANK_REASON Academic paper introducing a new method for 3D shape retrieval.

Read on arXiv cs.CV →

COVERAGE [2]

  1. arXiv cs.CV TIER_1 · Jiaxin Shi, Guofeng Zhang, Wufei Ma, Naifu Liang, Adam Kortylewski, Alan Vuile ·

    PASR: Pose-Aware 3D Shape Retrieval from Occluded Single Views

    arXiv:2604.22658v1 Announce Type: new Abstract: Single-view 3D shape retrieval is a fundamental yet challenging task that is increasingly important with the growth of available 3D data. Existing approaches largely fall into two categories: those using contrastive learning to map …

  2. arXiv cs.CV TIER_1 · Alan Vuile ·

    PASR: Pose-Aware 3D Shape Retrieval from Occluded Single Views

    Single-view 3D shape retrieval is a fundamental yet challenging task that is increasingly important with the growth of available 3D data. Existing approaches largely fall into two categories: those using contrastive learning to map point cloud features into existing vision-langua…