Sharper Frames, Worse Rooms: Viewpoint Coverage Dominates Image Quality in Feed-Forward 3D Reconstruction

Alibek Dadajonov; Hikmatillo Umarhojiyev

cs.CVOct 11, 2026

Sharper Frames, Worse Rooms: Viewpoint Coverage Dominates Image Quality in Feed-Forward 3D Reconstruction

Authors: Alibek Dadajonov, Hikmatillo Umarhojiyev

Organizations: Arbuz AI

Abstract

Grounded spatial reasoning about indoor scenes rests on a geometric reconstruction that the multimodal system does not itself produce. Feed-forward multi-view reconstruction models operate under a fixed view budget, making frame selection a critical upstream decision. We show that under this setting the conventional preference for sharp frames can be actively harmful. Across nine handheld room scans, selecting the sharpest 16 of 200 candidate frames is the worst strategy we test, reaching 48% room dimension error where uniform temporal sampling reaches 3%. We trace this failure to a coupling between image quality and camera motion: sharp frames tend to occur during slow motion and therefore concentrate the selection in a narrow angular region, leaving $192^\circ$ of the room unobserved in the case we illustrate and over $300^\circ$ in the worst cells. A $2\times2$ factorial experiment that independently manipulates viewpoint spread and per-frame sharpness shows that spread has roughly three times the main effect of quality, and that blurry but well spread selections outperform sharp but clustered selections in 8 of 9 captures. The advantage also diminishes as the frame budget increases and angular coverage saturates, consistent with this mechanism. Finally, we ask whether the failure is detectable at inference: the reconstruction model's multi-view consistency score increases with reconstruction error, while a vision-language model given the same frames reports confidence unrelated to its own spatial estimation error. Under small view budgets, viewpoint coverage is a more important constraint than per-frame image quality, and neither uncertainty signal available at inference reveals this upstream failure.

Figures & tables

capture	measured (m)	dense reference (m)	relative error
bedroom	$3.39\times 3.39$	$3.94\times 3.63$	11.8%
work_office	$3.90\times 2.62$	$4.22\times 2.83$	8.1%
home_office	$4.72\times 3.72$	$4.04\times 3.41$	11.4%

Table 1: Hand measured dimensions against the dense 200-frame reference, for the three captures we could physically access. Mean absolute error is 10% with no consistent direction: the reference overestimates both dimensions of bedroom and work_office and underestimates both dimensions of home_office. Room shape is nonetheless recovered, with the square room reconstructed as square. Note also that these three captures have the best self-consistency in the set (0.3 to 1.1%, Section 3.3 ) while showing 8 to 12% accuracy error, so internal agreement between dense reconstructions does not predict agreement with reality.

Figure 1: Sharpness ranked selection against uniform sampling on one capture ( $K{=}16$ ). Top: viewing directions of selected frames (colour) over the yaw distribution of all 200 candidates (grey). Bottom: reconstructed point cloud from above, with the fitted rectangle (solid) and the dense 200-frame reference (dashed), registered to a common centre. Rectangles use 5 to 95 percentile extents, so sparse peripheral points fall outside them.

	Relative error		Degenerate
	high quality	low quality	high quality	low quality
spread	$0.110\pm 0.077$	$0.154\pm 0.131$	0.000	0.025
clustered	$0.324\pm 0.152$	$0.435\pm 0.166$	0.086	0.395

Table 2:

2\times 2

factorial at

K{=}16

, marginalised over view ordering. Means and standard deviations over 81 reconstructions per cell from nine captures; inference treats the capture as the unit (

n{=}9

Figure 2: The factorial contrast on one capture: clustered-high against spread-high, identical mean sharpness (0.22), differing only in spread. Same conventions as Figure 1 .

$K$	8	16	32	48
clustered error	0.421	0.316	0.268	0.192
spread error	0.160	0.104	0.078	0.087
gap	0.262	0.212	0.190	0.105
coverage achieved (spread)	0.222	0.444	0.889	1.000
degeneracy, clustered-low	0.500	0.417	0.167	0.000

Table 3: Budget sweep over the four full sweep captures available when it was run, three seeds each, so twelve reconstructions per cell.

K{=}12

and

K{=}24

are omitted for space and interpolate smoothly.

	relative error	multi-view confidence
spread, high quality	0.110	0.444
spread, low quality	0.154	0.363
clustered, high quality	0.324	0.669
clustered, low quality	0.435	0.674

Table 4: Multi-view consistency score against error, means over 81 reconstructions per cell. The ordering is inverted throughout.

Figure 3: Neither downstream signal tracks the upstream failure. (a) Per capture and strategy, reconstruction error against the VLM’s room-dimension error: the selections that break the geometry are not the ones the VLM gets wrong. (b) Error distribution at each of the four confidence values the model actually reports, with individual queries jittered over the boxes. Of 189 queries, 131 return a confidence of exactly 0.30 and the 0.40 bin contains only eight, so the outer bins should not be over-read. Colour denotes capture. Section A gives the query text.

Explore similar work

Sep 21, 2026

When Wider Views Fail: Stress-Testing Feed-Forward 3D Reconstruction

Feed-forward 3D reconstruction models enable efficient geometry estimation from sparse images, but their pretrained nature can make them vulnerable to distribution shifts beyond their training data. Identifying these failure modes is important for understanding when such models can be reliably deployed in unconstrained imaging settings. We investigate viewpoint variation as a controlled distribution shift by varying the angular span of sparse image inputs while keeping the input budget fixed. Across multiple feed-forward reconstruction models, we observe substantial degradation as viewpoint span increases, with wide spans producing both incomplete surface coverage and geometry unsupported by the observed imagery. These results reveal that viewpoint variation can induce failure modes beyond conventional reconstruction incompleteness, highlighting the need to evaluate pretrained feed-forward models under distribution shifts that challenge their learned geometric priors.

Daisy Li, Kyle Gao, Quanyun Wu +3

University of Waterloo · Aalto University

Distribution Shift Robustness 3D Reconstruction Sparse-View 3D Reconstruction Feed-Forward 3D Reconstruction

Ask PDF

Aug 20, 2026

Sparse Light Field Sampling Improves Casual 3D and 4D Reconstruction

Many consumer smartphones, stereo cameras, and light field cameras record multiple synchronized viewpoints in a single exposure event. However, novel view synthesis pipelines commonly use only a monocular stream and rely on camera motion or learned priors to obtain angular coverage. In this paper, we ask: why do we use only one viewpoint? We analyze sensor-limited multi-view, where one sensor trades off spatial and angular resolution, and exposure-limited multi-view, where multiple sensors on one commodity device observe each event simultaneously. We introduce a new dataset incorporating three types of commodity multi-view cameras, and evaluate sparse-view 3DGS and 4DGS baselines measuring reconstruction quality as a function of number of exposures and angle between extreme views. Our results demonstrate that using multiple cameras, even with a low baseline, significantly improves reconstruction quality in single-shot, few-shot, and casual video settings. In addition, under a fixed sensor budget, angular sampling improves reconstruction when exposures are scarce despite lower spatial resolution. The gains are most pronounced for single-shot and dynamic scenes, where a stationary monocular camera lacks the angular diversity to recover scene geometry and motion.

Shamus Li, Ruiming Cao, Laura Waller +2

Cornell University, USA · Adobe, USA · University of California, Berkeley, USA +1

3D Reconstruction Sparse-View 3D Reconstruction 4D Reconstruction Dynamic Scene Reconstruction 3D Gaussian Splatting

Ask PDF

Apr 6, 2026

Coverage Optimization for Camera View Selection

What makes a good viewpoint? The quality of the data used to learn 3D reconstructions is crucial for enabling efficient and accurate scene modeling. We study the active view selection problem and develop a principled analysis that yields a simple and interpretable criterion for selecting informative camera poses. Our key insight is that informative views can be obtained by minimizing a tractable approximation of the Fisher Information Gain, which reduces to favoring viewpoints that cover geometry that has been insufficiently observed by past cameras. This leads to a lightweight coverage-based view selection metric that avoids expensive transmittance estimation and is robust to noise and training dynamics. We call this metric COVER (Camera Optimization for View Exploration and Reconstruction). We integrate our method into the Nerfstudio framework and evaluate it on real datasets within fixed and embodied data acquisition scenarios. Across multiple datasets and radiance-field baselines, our method consistently improves reconstruction quality compared to state-of-the-art active view selection methods. Additional visualizations and our Nerfstudio package can be found at https://chengine.github.io/nbv_gym/.

Timothy Chen, Adam Dai, Maximilian Adang +2

Stanford University

Active View Selection Next-Best-View Planning 3D Reconstruction Neural Radiance Fields

Ask PDF