Reconstructing a coherent 3D scene from uncurated image collections remains challenging due to the presence of irrelevant or non-overlapping views. Recent work has shown that feed-forward reconstruction models contain internal view-consistency signals that support distractor rejection. However, reference-conditioned filtering relies on a curated clean reference, which is not guaranteed in practical image collections. To address this limitation, we introduce FreeView, a training-free framework that jointly selects a filtering reference and rejects distractor views from an uncurated collection. FreeView leverages collection-level consensus to identify a representative view that serves as a reliable reference for filtering. It then adapts view rejection to each input collection by combining complementary visual and geometric evidence. This approach preserves useful views under substantial appearance and viewpoint variations while rejecting distractors. FreeView integrates with both VGGT and pi^3, removing the need for a manually curated reference or manual calibration of view rejection. Extensive experiments across multiple datasets demonstrate high clean-view retention and distractor rejection rates, leading to reliable 3D reconstruction from uncurated views.
Figures & tables
Fig. 2 : Overview of FreeView. Given an uncurated image collection, Consensus Reference Selection identifies a representative view through agreement among related images, establishing a reference for filtering without prior knowledge of which views are clean. Adaptive Dual-Cue Rejection combines complementary visual and geometric evidence to adapt view selection to each collection. Retaining views supported by either cue preserves useful observations under appearance and viewpoint variations, while rejecting views unsupported by both suppresses distractors. The retained views are then processed by VGGT or π3 for final reconstruction.
TABLE I : View filtering and pose estimation across distractor levels. For the same mixed image collections, reference-conditioned methods are evaluated with either a clean or a distractor reference to assess sensitivity to reference choice, while FreeView selects its reference automatically. CK/DR denote clean-view retention/distractor rejection; T/R denote translation/rotation recall over all clean views, counting rejected views as failures. For π3 , the reference is used only for filtering. Bold / underline indicate the best/second-best non-oracle results per backbone. † denotes DINOv2 features from the VGGT encoder.
CRS
ADCR
CK ↑
DR ↑
T@2/5/10% ↑
R@2/5/10 ∘↑
Phototourism
✗
✗
0.1
78.9
0.0 / 0.0 / 0.0
0.0 / 0.0 / 0.0
✓
✗
100.0
99.0
52.8 / 81.7 / 93.3
97.2 / 100.0 / 100.0
✓
✓
100.0
99.7
53.2 / 82.0 / 93.8
97.3 / 100.0 / 100.0
ETH3D
✗
✗
0.7
86.4
0.4 / 0.4 / 0.4
0.0 / 0.0 / 0.0
TABLE II : Ablation study on Consensus Reference Selection (CRS) and Adaptive Dual-Cue Rejection (ADCR).
Fig. 3 : Effect of neighborhood size on reference selection and view filtering on Phototourism under the Large setting, with 30 clean views and 50 distractors.
Filtering Cue
Fusion
CK ↑
DR ↑
T@2/5/10% ↑
R@2/5/10 ∘↑
DINOv2
VGGT Feat.
Phototourism
✓
✗
–
98.9
100.0
52.4 / 81.1 / 94.2
96.2 / 98.9 / 98.9
✗
✓
–
91.7
99.7
47.2 / 75.3 / 85.4
84.4 / 91.7 / 91.7
✓
✓
OR
85.0
100.0
42.9 / 70.0 / 80.3
77.8 / 85.0 / 85.0
✓
✓
Avg
62.8
100.0
40.4 / 56.7 / 62.2
56.1 / 62.8 / 62.8
TABLE III : Comparison of view filtering cues and fusion strategies.
Fig. 4 : Qualitative comparison of view filtering and 3D reconstruction.
Current end-to-end multi-view 3D reconstruction methods achieve impressive results, but rely on a restrictive static assumption: the scenes is entire distractor-free with perfect cross-view geometry. This reliance on idealized inputs causes even the most advanced methods to fail in real-world settings, where transient distractors and occlusions present. To address this, we propose Visual Geometry Transformer in the Wild (VGTW), an end-to-end framework for robust reconstruction from inconsistent views. At its core, we isolate and suppress distractor-affected regions while preserving the consistent components across views. Specifically, we introduce a Distractor-aware Training (DAT) strategy that separates clean features from distractor-contaminated ones in the attention mechanism while enforcing feature consistency across images. To enable this, we train the model with an auxiliary mask prediction head, using supervision from a new dataset we collected with pixel-level distractor masks. The resulting VGTW model is a feed-forward network that directly outputs clean, distractor-free point clouds. Remarkably, it requires no additional 3D supervision, remains computationally efficient, and is compatible with existing pipelines. Extensive experiments validate our approach, demonstrating state-of-the-art performance and robust generalization in diverse, real-world scenarios.
Tianbo Pan, Xingyi Yang, Shizun Wang +1
National University of Singapore · The Hong Kong Polytechnic University
Feed-forward 3D Gaussian Splatting reconstructs an explicit Gaussian representation from multiple input images in one network execution, making 3D reconstruction increasingly accessible for casual captures. However, such captures frequently contain transient objects that appear in only a subset of the views. Such content can be encoded into the per-view Gaussians associated with the inputs that observe it and remain in the combined representation despite being observed by no other input. As a result, it may produce blurred, duplicated, or floating artifacts in novel views. We introduce a training-free filtering procedure that exploits this per-view prediction structure. For each input, we exclude its associated Gaussians and render the same camera using the remaining representation, revealing content that is inconsistent with the other inputs. Feature similarity forms candidate regions, and rendering-based verification retains only candidates whose removal reduces reconstruction error in the other input views. The procedure operates on a single frozen prediction without retraining or scene-specific optimization. Across three reconstruction models and two distractor benchmarks, it consistently improves novel-view quality with varying numbers of input views. On clean scenes, evaluations across four models show that the original reconstructions are largely preserved.
Sparse-view 3D reconstruction is essential for modeling scenes from casual captures, but remain challenging for non-generative reconstruction. Existing diffusion-based approaches mitigates this issues by synthesizing novel views, but they often condition on only one or two capture frames, which restricts geometric consistency and limits scalability to large or diverse scenes. We propose AnyRecon, a scalable framework for reconstruction from arbitrary and unordered sparse inputs that preserves explicit geometric control while supporting flexible conditioning cardinality. To support long-range conditioning, our method constructs a persistent global scene memory via a prepended capture view cache, and removes temporal compression to maintain frame-level correspondence under large viewpoint changes. Beyond better generative model, we also find that the interplay between generation and reconstruction is crucial for large-scale 3D scenes. Thus, we introduce a geometry-aware conditioning strategy that couples generation and reconstruction through an explicit 3D geometric memory and geometry-driven capture-view retrieval. To ensure efficiency, we combine 4-step diffusion distillation with context-window sparse attention to reduce quadratic complexity. Extensive experiments demonstrate robust and scalable reconstruction across irregular inputs, large viewpoint gaps, and long trajectories.
Yutian Chen, Shi Guo, Renbiao Jin +7
Shanghai AI Lab · Shanghai Jiao Tong University · The University of Hong Kong +2