Estimating image depth using shape collections

Su, Hao; Huang, Qixing; Mitra, Niloy J.; Li, Yangyan; Guibas, Leonidas J.

doi:10.1145/2601097.2601159

Cited by 75 publications

(68 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We compared the camera poses produced by our approach, which is based on joint estimation over an image collection, to camera poses produced by the approach of Su et al [2014], which estimates a camera pose for each image separately. Accuracy was measured by angular deviations from ground truth.…”

Section: Discussionmentioning

confidence: 99%

“…Accuracy of pixel-level correspondences is measured by pixel distance from ground truth. We compare the accuracy of the presented approach to the accuracy of correspondences estimated by the approach of Su et al [2014]. Cumulative error distributions are shown in Figure 7.…”

Section: Discussionmentioning

confidence: 99%

“…This baseline represents the performance of an oracle that retrieves the best preexisting model. In addition, we measured the accuracy of depth maps synthesized by the approach of Su et al [2014]; the output of this approach is a point cloud that models only one aspect of the object, but its accuracy can still be estimated by computing the average Euclidean distance to the ground truth shape.…”

Section: Discussionmentioning

confidence: 99%

“…Estimation of three-dimensional layout from a single image can be performed using projective geometry techniques [Criminisi et al 2000] or by data-driven approaches [Fouhey et al 2013;Su et al 2014;Eigen et al 2014]. In contrast to these techniques, our approach reconstructs complete 3D models, including surfaces that are occluded in input images.…”

Section: Related Workmentioning

confidence: 99%

“…A natural approach to estimating the pose of an image I ∈ I is to retrieve the most similar rendered images and use their known camera poses to estimate the pose of I [Aubry et al 2014;Lim et al 2014;Su et al 2014]. We found that the results produced by this approach are noisy, as illustrated in Figure 3.…”

Section: Camera Pose Estimationmentioning

confidence: 99%

See 4 more Smart Citations

Single-view reconstruction via joint analysis of image and shape collections

2015

Self Cite

View full text Add to dashboard Cite

Web image searchReconstructed 3D modelsFigure 1: Our approach reconstructs objects depicted in images, even if each object is only shown in a single image. Left: Web image search for "windsor chair", first page results. Right: 3D models automatically generated by our approach for these images. AbstractWe present an approach to automatic 3D reconstruction of objects depicted in Web images. The approach reconstructs objects from single views. The key idea is to jointly analyze a collection of images of different objects along with a smaller collection of existing 3D models. The images are analyzed and reconstructed together. Joint analysis regularizes the formulated optimization problems, stabilizes correspondence estimation, and leads to reasonable reproduction of object appearance without traditional multi-view cues.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Camera Pose Estimationmentioning

confidence: 99%

See 3 more Smart Citations

Single-view reconstruction via joint analysis of image and shape collections

2015

Self Cite

View full text Add to dashboard Cite

show abstract

Evaluating the accuracy of unmanned aerial systems to quantify glacial ice habitats of harbor seals in Alaska

et al. 2022

View full text Add to dashboard Cite

Long-term monitoring programs to evaluate climate-driven changes to tidewater glaciers, an important habitat for harbor seals (Phoca vitulina) in Alaska, are primarily carried out by costly, weather-dependent aerial surveys from fixed-winged aircraft. Unmanned aerial systems (UAS) can be an alternative cost-effective solution for gathering image data to quantify, monitor, and manage these habitats. However, there is a paucity of information related to the accuracy of using imagery collected by UAS for purposes of measuring floating icebergs. We evaluated the accuracy of using a UAS with a built-in 20-megapixel (MP) camera as well as a consumer-grade digital 16-MP camera to capture images of floating and stationary icebergs for the purpose of collecting vertical height measurements. Images (n = 869) were captured of simulated icebergs (cuboidal foam boxes, Cb) (n = 5) and real icebergs (n = 5) that were either grounded or floating. The mean error ratios (Ers) obtained were less than 10% and derived by comparing the mean calculated measurements of heights of Cb obtained from images captured by UAS with the physically measured heights of these Cb. The mean Er for height measurements of grounded icebergs (n = 4) and one floating iceberg was also less than 10%. Within an object-image distance range of 6-25 m, the cameras captured images that were suitable to accurately calculate the heights of floating and grounded objects, and drift or uncontrolled movement of the UAS caused by wind or temporary loss of GPS did not have any effect on measurement error. Our study provides substantial evidence of the accuracy associated with using images captured by UAS for measuring dimensions of structures positioned on either water or land surfaces. Ultimately, accurate surveys of glacial ice used by harbor seals will improve our understanding of the role of decreasing habitat in explaining population variability between different tidewater glaciers. K E Y W O R D S drones, harbor seal, icebergs, tidewater glaciers, unmanned aerial systems † Deceased.

show abstract

Pixel2Mesh: Generating 3D Mesh Models from Single RGB Images

Wang

Zhang

et al. 2018

Computer Vision – ECCV 2018

994

1,147

View full text Add to dashboard Cite

We propose an end-to-end deep learning architecture that produces a 3D shape in triangular mesh from a single color image. Limited by the nature of deep neural network, previous methods usually represent a 3D shape in volume or point cloud, and it is non-trivial to convert them to the more ready-to-use mesh model. Unlike the existing methods, our network represents 3D mesh in a graph-based convolutional neural network and produces correct geometry by progressively deforming an ellipsoid, leveraging perceptual features extracted from the input image. We adopt a coarse-to-fine strategy to make the whole deformation procedure stable, and define various of mesh related losses to capture properties of different levels to guarantee visually appealing and physically accurate 3D geometry. Extensive experiments show that our method not only qualitatively produces mesh model with better details, but also achieves higher 3D shape estimation accuracy compared to the state-of-the-art.

show abstract

Estimating image depth using shape collections

Cited by 75 publications

References 44 publications

Single-view reconstruction via joint analysis of image and shape collections

Single-view reconstruction via joint analysis of image and shape collections

Evaluating the accuracy of unmanned aerial systems to quantify glacial ice habitats of harbor seals in Alaska

Pixel2Mesh: Generating 3D Mesh Models from Single RGB Images

Contact Info

Product

Resources

About