Gate3DGen: Selective Pre-processing Model That Purifies Real-World Images for 3D Generation

Forum topic · 小凯 · 2026-10-24

Summary

Gate3DGen (arXiv: 2610.31649) is a selective pre-processing model designed to improve single-image 3D generation on real-world photographs, which typically suffer from uneven illumination, blur, and occlusions. Instead of retraining existing 3D generators, Gate3DGen acts as a gatekeeper on their image inputs. A degradation-aware router first classifies each input: clean images are bypassed untouched, while degraded images are routed to a 3D-oriented one-step purifier. This purifier is conditioned on router-extracted features and trained with spatial losses that preserve object identity, contours, and texture cues needed for accurate 3D reconstruction. For paired supervision, the authors built a synthetic data engine rendering clean views of 3D assets and compositing realistic degradations. On synthetic benchmarks, the purifier reduces TRELLIS.2 mesh distance by 61.5% relative to degraded inputs, averaging only 2.3 seconds per image. VLM-based evaluation and human preference studies on real photographs further confirm its identity preservation and degradation correction capabilities. The work is authored by Xiaolin Li, Zihao Huang, Fanpeng Meng, Chen Yang, Guanjun Wu, Kai Zhang, Jiemin Fang, and Qi Tian.

Overview

Field: Computer Vision
Authors: Xiaolin Li, Zihao Huang, Fanpeng Meng, Chen Yang, Guanjun Wu, Kai Zhang, Jiemin Fang, Qi Tian
Published: 2026-10-22
arXiv: 2610.31649

Summary

Single-image 3D generation has achieved high fidelity on curated benchmarks, yet degrades sharply on real-world images with uneven illumination, blur, and occlusions. Purifying such inputs requires deciding when to intervene, preserving the visual cues needed for 3D generation, while leaving clean inputs untouched.

The authors propose Gate3DGen, a selective pre-processing model that acts as a gatekeeper for the image inputs of existing 3D generators, without requiring them to be retrained.

A degradation-aware router bypasses inputs classified as clean and extracts features to guide purification of the remaining inputs.
Conditioned on these features, a 3D-oriented one-step purifier purifies degraded inputs using spatial losses that encourage preservation of object identity, contours, and textural cues.
To provide paired supervision, the authors built a synthetic data engine that renders clean views of 3D assets and synthesizes composite degradations.

Results

On a synthetic benchmark, the purifier reduces TRELLIS.2 mesh distance by 61.5% relative to degraded inputs, averaging only 2.3 seconds per image.
VLM evaluation and human preference studies on real photographs further support its identity-preservation and degradation-correction capabilities.

Original Abstract (excerpt)

> Single-image 3D generation has achieved high fidelity on curated benchmarks, yet degrades sharply on real-world images with uneven illumination, blur, and occlusions. We introduce Gate3DGen, a selective pre-processing model that gates image inputs to existing 3D generators without retraining them. A degradation-aware router bypasses inputs classified as clean and extracts features to guide purification of the remaining inputs. Conditioned on these features, a 3D-oriented one-step purifier purifies degraded inputs with spatial losses that encourage preservation of object identity, contours, and textural cues. On the synthetic benchmark, the purifier reduces TRELLIS.2 mesh distance by 61.5% relative to degraded inputs. VLM evaluation and human preferences on real photographs further support ...

*Auto-collected on 2026-10-24*

Gate3DGen: Selective Pre-processing Model That Purifies Real-World Images for 3D Generation

Summary

Overview

Summary

Results

Original Abstract (excerpt)

Tags