Efficient Depth-Guided Urban View Synthesis

Andreas Geiger; Bingbing Liu; Dongfeng Bai; Jiaxin Huang; Sheng Miao; Weichao Qiu; Yiyi Liao

RECORD

Efficient Depth-Guided Urban View Synthesis

Sheng Miao, Jiaxin Huang, Dongfeng Bai, Weichao Qiu, Bingbing Liu, Andreas Geiger, Yiyi Liao

Computer Vision and Pattern Recognition · Submitted 17 Jul 2024

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim Moreover, our results indicate that EDUS achieves state-of-the-art performance in sparse view settings when combined with fast test-time optimization.

how it is stated · hedged possibility 0.85 read the record →

Comment (0) Read on arXiv ↗

Cite ▾

arXiv ID2407.12395 (version 2)

Pith IDW2RCH6PL, Pith's permanent record of this paper and its review

CitationAndreas Geiger, Bingbing Liu, Dongfeng Bai, Jiaxin Huang, Sheng Miao, Weichao Qiu, Yiyi Liao. "Efficient Depth-Guided Urban View Synthesis." arXiv:2407.12395 (2024). https://pith.science/paper/W2RCH6PL

Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 5,586 input tokens.

What the abstract itself states, as probabilities: claims first 0.06 · states limitation 0.16 · human participants 0.02 · controlled comparison 0.37 · randomized assignment 0.01 · states quantitative result 0.05 · compares with prior results 0.87 · mentions code or data availability 0.02.

The abstract, sentence by sentence

arXiv listing abstract v2. Results and the main claim are underlined; show every role.

Recent advances in implicit scene representation enable high-fidelity street view novel view synthesis. However, existing methods optimize a neural radiance field for each scene, relying heavily on dense training images and extensive computation resources. To mitigate this shortcoming, we introduce a new method called Efficient Depth-Guided Urban View Synthesis (EDUS) for fast feed-forward inference and efficient per-scene fine-tuning. Different from prior generalizable methods that infer geometry based on feature matching, EDUS leverages noisy predicted geometric priors as guidance to enable generalizable urban view synthesis from sparse input images. The geometric priors allow us to apply our generalizable model directly in the 3D space, gaining robustness across various sparsity levels. Through comprehensive experiments on the KITTI-360 and Waymo datasets, we demonstrate promising generalization abilities on novel street scenes. Moreover, our results indicate that EDUS achieves state-of-the-art performance in sparse view settings when combined with fast test-time optimization.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#	sentence	role	the authors' own work	states the main conclusion	strength
s0	Recent advances in implicit scene representation enable high-fidelity street view novel view synthesis.	background 1.00	0.04	0.03	asserted 1.00
s1	However, existing methods optimize a neural radiance field for each scene, relying heavily on dense training images and extensive computation resources.	gap 0.65	0.05	0.02	asserted 1.00
s2	To mitigate this shortcoming, we introduce a new method called Efficient Depth-Guided Urban View Synthesis (EDUS) for fast feed-forward inference and efficient per-scene fine-tuning.	contribution 1.00	0.87	0.12	asserted 0.99
s3	Different from prior generalizable methods that infer geometry based on feature matching, EDUS leverages noisy predicted geometric priors as guidance to enable generalizable urban view synthesis from sparse input images.	method 0.89	0.60	0.10	asserted 1.00
s4	The geometric priors allow us to apply our generalizable model directly in the 3D space, gaining robustness across various sparsity levels.	method 0.74	0.77	0.07	asserted 0.98
s5	Through comprehensive experiments on the KITTI-360 and Waymo datasets, we demonstrate promising generalization abilities on novel street scenes.	result 0.91	0.97	0.24	asserted 0.96
s6	Moreover, our results indicate that EDUS achieves state-of-the-art performance in sparse view settings when combined with fast test-time optimization.	result 1.00	0.96	0.57	hedged possibility 0.85

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.97.

What was examined, and what was not

sources read

arXiv listing abstract v2

not examined

full text
figures and tables
references and citing sentences
relations to other papers

levels run

P abstract

not run

S sections
T body sentences
C citations
R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper

Original abstract

Recent advances in implicit scene representation enable high-fidelity street view novel view synthesis. However, existing methods optimize a neural radiance field for each scene, relying heavily on dense training images and extensive computation resources. To mitigate this shortcoming, we introduce a new method called Efficient Depth-Guided Urban View Synthesis (EDUS) for fast feed-forward inference and efficient per-scene fine-tuning. Different from prior generalizable methods that infer geometry based on feature matching, EDUS leverages noisy predicted geometric priors as guidance to enable generalizable urban view synthesis from sparse input images. The geometric priors allow us to apply our generalizable model directly in the 3D space, gaining robustness across various sparsity levels. Through comprehensive experiments on the KITTI-360 and Waymo datasets, we demonstrate promising generalization abilities on novel street scenes. Moreover, our results indicate that EDUS achieves state-of-the-art performance in sparse view settings when combined with fast test-time optimization.

Human review

How Pith checks papers → · Full audit of this paper →

How the literature uses it

3 arXiv papers cite this in 7 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it · 2 sentences without a confident label

Contests it 1 paper

PointNeRF [ref] and EDUS [ref] learn implicit radiance fields from dense input point clouds, limiting their practical applicability due to slow rendering.

EVolSplat: Efficient Volume-based Gaussian Splatting for Urban View Synthesis · Related Work · contest 0.76

Compares with it 1 paper

Baselines We compare with several representative feed-forward methods, including MVSNeRF [ref] , MuRF [ref] , EDUS [ref] , MVSplat [ref] and PixelSplat [ref] .

EVolSplat: Efficient Volume-based Gaussian Splatting for Urban View Synthesis · Experimental Setup · 1.00

Cites it as background 2 papers

Recent advances in neural rendering, from NeRF [ref] to 3D-GS [ref] , have revolutionized static scene reconstruction and novel view synthesis.

Vivid4D: Improving 4D Reconstruction from Monocular Video by Video Inpainting · Related Work · 0.99
These methods can be categorized into four classes: point-based [ref] , mesh-based [ref] , NeRF-based [ref] and 3DGS-based [ref] .

HUGSIM: A Real-Time, Photo-Realistic and Closed-Loop Simulator for Autonomous Driving · Urban Scene Reconstruction · 1.00

What it cites · 3 sentences without a confident label

Uses the method or tool of 6 papers

To maintain parity in our evaluation, we furnish the depth maps utilized in our method to train depth-supervision approaches [ref] , and employ our accumulated point cloud as the initialization for 3DGS [ref] .

SparseNeRF: Distilling Depth Ranking for Few-shot Novel View Synthesis · Experiment · 0.84
Hierarchical Sampling For rendering a ray, we follow NeuS [ref] to perform hierarchical sampling during training and inference.

NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstruction · Implementation details · 1.00
Instead, we draw inspiration from [ref] to employ a 3D Spatially-Adaptive Normalization convolutional neural network (SPADE CNN) to modulate feature volume generation at multiple scales, which is a popular strategy in semantic image synthesis.

UrbanGIRAFFE: Representing Urban Scenes as Compositional Generative Neural Feature Fields · Depth Estimation: · 0.99

Compares with 2 papers

Secondly, we compare with test-time optimization methods, including methods for sparse view settings: MixNeRF [ref] , SparseNeRF [ref] , DS-NeRF [ref] , and one taking point clouds as input: 3DGS [ref] .

SparseNeRF: Distilling Depth Ranking for Few-shot Novel View Synthesis · Experiment · 0.95
Firstly, we compare our model with generalizable approaches: IBRNet [ref] , MVSNeRF [ref] , NeO 360 [ref] and MuRF [ref] .

MuRF: Multi-Baseline Radiance Fields · Experiment · 1.00

Cites as support 1 paper

As suggested in the current work MuRF [ref] , these methods struggle to generalize well to large baselines.

MuRF: Multi-Baseline Radiance Fields · Related Work · 0.97

Cites as background 8 papers

Several recent methods address the sparse view setting: One line of works [ref] exploits external geometric priors or regularization terms, such as depth maps, geometric continuity, and color cues, to enhance model performance.

SparseNeRF: Distilling Depth Ranking for Few-shot Novel View Synthesis · Introduction · 0.83
[ref] learn a scene starting from a coarse point cloud, which can be obtained from multi-view stereo techniques [ref] or learning-based methods [ref] by unprojecting depth maps.

Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data · Related Work · 0.83
Generalizable NeRF approaches are also intensively studied in the literature [ref] .

MuRF: Multi-Baseline Radiance Fields · Related Work · 0.81

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.