High-Resolution Stereo Datasets with Subpixel-Accurate Ground Truth

By Daniel Scharstein, Heiko Hirschmüller, York Kitajima, Greg Krathwohl, Nera Nesic, Xi Wang, Porter Westling. 2014

1,491 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Geology

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 Geology

Citations in the corpus

SafetyNet: Detecting and Rejecting Adversarial Examples Robustly (2017, methodology): ““Regression” methods used only in test are: all zero depth values; nearest neighbor down-sample and up-sampled images and depths; low quality JPEG compressed images and depths; Middlebury stereo RGBD dataset [28] and Sintel RGBD dataset [1](which should be classified “fake” because they are renderin”
IRS: A Large Synthetic Indoor Robotics Stereo Dataset for Disparity and Surface Normal Estimation (2019, background): “Existing studies [10], [11], [12], [13] proposed stereo datasets collected by real sensing hardware, which contributed a lot to stereo vision research.”
A Large Dataset to Train Convolutional Networks for Disparity, Optical Flow, and Scene Flow Estimation (2015, background): “An advantage of the stereo dataset is the availability of relevant real scenes, especially in the latest high-resolution version from 2014 [19].”
Playing for Benchmarks (2017, background): “The Middlebury benchmarks for stereo, multi-view stereo, and optical ﬂow played a pivotal role in the development of low-level vision algorithms [4, 57, 58, 59].”
Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches (2015, methodology): “The data sets were published in ﬁve separate works in the years 2001, 2003, 2005, 2006, and 2014 (Scharstein and Szeliski, 2002, 2003; Scharstein and Pal, 2007; Hirschm¨uller and Scharstein, 2007; Scharstein et al., 2014).”
Robust stereo matching with surface normal prediction (2017, methodology): “We evaluate our method on both indoor stereo data captured by ourselves and the public Middlebury stereo dataset [29], [30], [18], [28].”
Fast Optical Flow Using Dense Inverse Search (2016, methodology): “We remove the vertical degree of freedom from our method, and evaluate on the Middlebury dataset for depth from stereo [7].”
Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (2024, neutral): “Using the pretrained weights, we train each backbone at 384 × 384 resolution across five RGB-D datasets (Keystone, HRWSI, RedWeb, TartanAir, and Hypersim) and evaluate their performance in terms of metric depth accuracy across multiple datasets, including Booster, Hypersim, Middlebury, and NYUv2, ut”
Cycle-Dehaze: Enhanced CycleGAN for Single Image Dehazing (2018, methodology): “We have chosen NYU-Depth [28], which is considerably larger scale than Middlebury [26] part.”
Deep Depth Completion of a Single RGB-D Image (2018, methodology): “n them to images captured simultaneously with a higher cost depth sensor. This approach is costly and time-consuming – the largest public datasets of this type cover a handful of indoor scenes (e.g., [57,62,75]). Instead, to create our dataset, we utilize existing surface meshes reconstructed from m”
Hierarchical Neural Architecture Search for Deep Stereo Matching (2020, methodology): “We then conduct the architecture evaluation on KITTI 2012 [29], KITTI 2015 [30] and Middlebury 2014 [31] benchmarks by inheriting the searched architecture from SceneFlow dataset.”
Hierarchical Deep Stereo Matching on High-Resolution Images (2019, methodology): “During training, we augment Middlebury, KITTI-15, ETH3D [20, 21, 26] and HR-VS to the same size as Sceneflow, resulting in around 170k training samples.”
Practical Stereo Matching via Cascaded Recurrent Network with Adaptive Correlation (2022, background): “Middlebury 2014 [33] provides 23 high-resolution image pairs under different lighting environments.”
Learning for Disparity Estimation Through Feature Constancy (2017, neutral): “The four-step pipeline dominants existing stereo matching algorithms [24, 22, 9, 21], while each of its steps is important to the overall stereo matching performance.”
Shading-based refinement on volumetric signed distance functions (2015, background): “This is not only due to averaging in the SDF, but also due to the stereo reconstruction itself, which often requires strong regularization to ﬁnd image correspondences from scene texture [Seitz et al. 2006; Scharstein et al. 2014].”
Paris-Lille-3D: A large and high-quality ground-truth urban point cloud dataset for automatic segmentation and classification (2017, methodology): “• Stereovision dataset for depth map estimation: Middlebury Stereo Datasets3 (Scharstein et al., 2014).”
Semi-global Stereo Matching with Surface Orientation Priors (2017, methodology): “We start by evaluating the promise of SGM-P on a subset of the high-resolution stereo pairs from the training set of the Middlebury stereo evaluation v3 [38], for which highquality ground-truth disparities are available.”
A Dataset and Evaluation Methodology for Depth Estimation on 4D Light Fields (2016, background): “As in [19] algorithm results of the training scenes will be available for download in full resolution.”
Synthetic Data for Deep Learning (2019, background): “Middlebury 2014 2014 [525] 33 high-res stereo datasets”