Pixelwise View Selection for Unstructured Multi-View Stereo

By Johannes L. Schönberger, Enliang Zheng, Jan-Michael Frahm, Marc Pollefeys. 2016

2,858 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology, 3D Vision & Pose

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

In 3D Vision & Pose

Citations in the corpus

Multiview Compressive Coding for 3D Reconstruction (2023, methodology): “CO3D extracts point clouds from the videos via COLMAP [54, 55], which are inevitably noisy and are used to train our model.”
Visual Imitation Made Easy (2020, methodology): “ons’ taken by the human demonstrator is crucial. To address this, we recover 6DoF poses of the tool using Structure-from-Motion (SfM) reconstruction. Speciﬁcally, we use the publicly available COLMAP [44, 45] software for SfM. Once we have the end-effector pose p tfor every image I t, we extract the”
Learning 3D Object Categories by Looking Around Them (2017, methodology): “We extract this information using off-theshelf methods: the structure-from-motion (SFM) algorithm COLMAP for RGB sequences [33, 34], and an open-source implementation [31] of KinectFusion (KF) [24] for RGBD sequences.”
Cost Volume Pyramid Based Depth Inference for Multi-View Stereo (2019, neutral): “Overall ( mm ) G e o m e ti c Furu[11] 0.613 0.941 0.777 Tola[39] 0.342 1.190 0.766 Camp[3] 0.835 0.554 0.695 Gipuma[12] 0.283 0.873 0.578 Colmap[33, 34] 0.400 0.664 0.532 L ea r n i ng SurfaceNet[17] 0.450 1.040 0.745 MVSNet[42] 0.396 0.527 0.462 P-MVSNet[26] 0.406 0.434 0.420 R-MVSNet[43] 0.383…”
MonoSDF: Exploring Monocular Geometric Cues for Neural Implicit Surface Reconstruction (2022, methodology): “…a) state-of-the-art neural implicit surfaces methods: UNISURF [49], VolSDF [81], NeuS [76], and Manhattan-SDF [21]. b) Classic MVS methods: COLMAP [62] and a state-of-the-art commercial software (RealityCapture 2 ). c) TSDF-Fusion [12] with predicted monocular depth cues, where GT depth maps are…”
Non-Rigid Neural Radiance Fields: Reconstruction and Novel View Synthesis of a Dynamic Scene From Monocular Video (2020, methodology): “If the camera extrinsics and intrinsics are not given, we estimate them using the Structure-fromMotion (SfM) implementation of COLMAP [68, 69].”
DeepVoxels: Learning Persistent 3D Feature Embeddings (2018, background): “Model-Based Rendering Classic reconstruction approaches such as structure-from-motion exploit multi-view geometry [15, 50] to build a dense 3D point cloud of the imaged scene [46, 47, 49, 1, 12].”
NeRD: Neural Reflectance Decomposition from Image Collections (2020, methodology): “All NeRD requires as input is a set of images of an object with known camera pose (computed for e.g . using COLMAP [45, 46]), where each image is accompanied by a foreground segmentation mask.”
DeepMVS: Learning Multi-view Stereopsis (2018, methodology): “Recently, Sch ¨ onberger et al. present a MVS system — COLMAP [33] — that jointly estimates depth and surface normal, leverages photometric and geometric priors for pixelwise view selection, and uses geometric consistency for simultaneous reﬁnement.”
Neural Body: Implicit Neural Representations with Structured Latent Codes for Novel View Synthesis of Dynamic Humans (2020, methodology): “We test state-ofthe-art multi-view methods COLMAP [54, 55] and DVR [47] on the ZJU-MoCap dataset.”
Depth Anything 3: Recovering the Visual Space from Any Views (2025, positive): “Traditional systems [70, 71] decompose reconstruction into feature detection and matching, robust relative pose estimation, incremental or global SfM with bundle adjustment, and dense multi-view stereo for per-view depth and fused point clouds.”
Neural RGB-D Surface Reconstruction (2021, background): “[65] Johannes Lutz Schönberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm.”
Deep Stereo Using Adaptive Thin Volume Representation With Uncertainty Awareness (2019, background): “Multi-view stereo is a long-studied vision problem with many traditional approaches [44, 39, 33, 32, 26, 10, 8, 13, 43].”
DeMoN: Depth and Motion Network for Learning Monocular Stereo (2016, background): “SUN3D & RGBD • Depth from structured light sensor • Camera pose from SfM (SUN3D) or external tracking (RGBD) Scenes11 • Randomly generated scenes and objects from ShapeNet[2] MVS • Collection of Multi VIew Stereo datasets • Depth and camera poses from SfM pipelines[4,10,11,16] Blendswap • About 150 ”
MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motion (2024, methodology): “Existing methods, such as COLMAP (Sch ¨ onberger & Frahm, 2016), often struggle with complex camera trajectories or highly dynamic scenes, making it challenging to produce even pseudo ground truth data for training.”
Semantic Stereo for Incidental Satellite Images (2018, background): “One of the significant challenges in multi-view stereo is reliable image pair selection based on known imaging geometry [35] and temporal proximity [1].”