A Theory of Shape by Space Carving

By Kiriakos N. Kutulakos, Steven M. Seitz. 1999

How researchers see this paper

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

In 3D Vision & Pose

Citations in the corpus

SurfaceNet: An End-to-End 3D Neural Network for Multiview Stereopsis (2017, background): “While earlier works like space carving [14, 22] mainly use a volumetric representation, current state-of-the-art MVS methods focus on depth map fusion algorithms [27, 4, 9], which have been shown to be more competitive in handling large datasets in practice.”
Learning a Multi-View Stereo Machine (2017, background): “The underlying principle behind MVS is simple - a 3D point looks locally similar when projected to different viewpoints [29].”
Multi-view Supervision for Single-View Reconstruction via Differentiable Ray Consistency (2017, background): “When is a solid 3D shape consistent with a 2D image? If it is not, how do we change it to make it more so? One way this problem has been traditionally addressed is by space carving [20].”
Hierarchical Surface Prediction for 3D Object Reconstruction (2017, methodology): “The common methods can be broadly grouped as implicit volumetric reconstruction [7], [18], [19], [22], [38] or explicit mesh based [9], [12] approaches.”
3D-R2N2: A Unified Approach for Single and Multi-view 3D Object Reconstruction (2016, background): “In order to circumvent issues related to large baselines or non-Lambertian surfaces, 3D volumetric reconstruction methods such as space carving [12,13,14,15] and their probabilistic extensions [16] have become popular.”
Perspective Transformer Nets: Learning Single-View 3D Object Reconstruction without 3D Supervision (2016, methodology): “In other words, if all the generated silhouettes Ŝ match well with their corresponding ground truth silhouettes S for all j’s, then we hypothesize that the generated volume V̂ should be as good as one instance of visual hull equivalent class of the ground truth volume V [10].”
Synthesizing 3D Shapes via Modeling Multi-view Depth Maps and Silhouettes with Deep Generative Networks (2017, background): “This is made possible because, ﬁrst, a multi-view 2D representation still offers rich ﬂexibility and allows generating detailed 3D shapes [17, 14].”
Point-NeRF: Point-based Neural Radiance Fields (2022, background): “Multi-view 3D reconstruction has been extensively studied and addressed with a number of structure-from-motion [41, 46, 47] and multi-view stereo techniques [10, 14, 24, 42, 54].”
Editing Conditional Radiance Fields (2021, background): “Prior work reasons over an explicit [11, 15, 22, 72] or discrete volumetric [20, 33, 36, 40, 44, 55, 64, 65, 67, 68, 80, 82] representation of the underlying geometry.”
Multi-resolution real-time stereo on commodity graphics hardware (2003, methodology): “Hardware-based SSD aggregation Although the simple consistency measure might be sufficient in multi-view approaches [9, 16], it is necessary to use larger support regio n in the two-view stereo case.”
ShapeCodes: Self-supervised Feature Learning by Lifting Views to Viewgrids (2017, background): “Image-based models for object shape (implicitly) intersect silhouette images to carve a visual hull [42, 43].”
Advances in Computational Stereo (2003, methodology): “Kutulakos and Seitz [48] represent the scene as a volume and propose a space carving method to refine the surface.”
A quasi-dense approach to surface reconstruction from uncalibrated images (2005, methodology): “For only 2D images and camera geometry, the recent volumetric methods [9], [24], [25], [46], are the most general image-based approaches, but they are not robust enough.”
MVSNet: Depth Inference for Unstructured Multi-view Stereo (2018, methodology): “According to output representations, MVS methods can be categorized into 1) direct point cloud reconstructions [23,4], 2) volumetric reconstructions [24,25,17,18] and 3) depth map reconstructions [19,20,6,7].”