Beyond PASCAL: A benchmark for 3D object detection in the wild

By Yu Xiang, Roozbeh Mottaghi, Silvio Savarese. 2014

846 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Evaluation & Benchmarks, 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 Aerospace Engineering

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

In 3D Vision & Pose

Citations in the corpus

Hypersim: A Photorealistic Synthetic Dataset for Holistic Indoor Scene Understanding (2020, background): “our dataset aims to be as photorealistic as possible, thereby signiﬁcantly reducing this domain gap. Several datasets provide 3D CAD models that have been aligned to individual objects in real images [17,44,68,75, 76]. In these datasets, the CAD models may not be exactly aligned to each image, and m”
ShapeNet: An Information-Rich 3D Model Repository (2015, background): “…contain segmentations [2], correspondences [13, 12], hierarchies [19], symmetries [11], salient features [3], semantic segmentations and labels [36], alignments of 3D models with images [35], semantic ontologies [5], and other functional annotations — but again only for small size datasets.”
Learning Shape Priors for Single-View 3D Completion and Reconstruction (2018, methodology): “Answers are available in the footnote. researchers have made impressive progress on the these tasks [7,52,8], making use of gigantic 3D datasets [5,60,59].”
Pix3D: Dataset and Methods for Single-Image 3D Shape Modeling (2018, methodology): “Many of the aforementioned object datasets include annotations of object poses [34, 38, 47, 63, 64].”
MarrNet: 3D Shape Reconstruction via 2.5D Sketches (2017, methodology): “Data PASCAL 3D+ dataset [Xiang et al., 2014] provides (rough) 3D models for objects in real-life images.”
Multi-view Supervision for Single-View Reconstruction via Differentiable Ray Consistency (2017, methodology): “We use annotated pose (in PASCAL 3D [37]) and segmentation masks (from PASCAL VOC) as training signal for object reconstruction.”
3D-R2N2: A Unified Approach for Single and Multi-view 3D Object Reconstruction (2016, methodology): “PASCAL 3D: The PASCAL 3D dataset is composed of PASCAL 2012 detection images augmented with 3D CAD model alignment [45].”
Learning a Predictable and Generative Vector Representation for Objects (2016, background): “0 [37] dataset on all categories that overlap with our training categories (chair and sofa).”
Lifting Object Detection Datasets into 3D (2015, background): “…a variety of tasks: (1) to reconstruct from a single image, usually with some user interaction to initialize the viewpoint [31], (2) as a prior for reconstruction from multiple images [34] or from a depth map [35], or (3) for performing object detection and pose estimation [33] in a single image.”
Category-specific object reconstruction from a single image (2014, methodology): “Since ground truth 3D shapes are unavailable for PASCAL VOC and the other high profile detections datasets, we evaluated the expressiveness of our learned 3D models on the next best thing we managed to obtain: the PASCAL3D+ dataset[47] which has up to 10 3D CAD models for the rigid categories in PAS”
Single Image 3D Interpreter Network (2016, methodology): “VDPM [17] and DPM-VOC+VP [50] are trained on PASCAL VOC 2012, V&K [18] is trained on PASCAL 3D+, Su et al .”
Render for CNN: Viewpoint Estimation in Images Using CNNs Trained with Rendered 3D Model Views (2015, methodology): “Here, we compare these methods again, using detection windows from VDPM, i.e., to estimate the viewpoint of objects in detection windows from VDPM.”
Online Meta-Learning (2019, neutral): “These tasks include synthetic modifications of the MNIST dataset, pose detection with synthetic images based on PASCAL3D+ models (Xiang et al., 2014), and realistic online image classification experiments with the CIFAR-100 dataset.”
Viewpoints and keypoints (2014, methodology): “[37] introduce a separate DPM component corresponding to each viewpoint.”
Adapting Deep Visuomotor Representations with Weak Pairwise Constraints (2015, methodology): “mance over the real-only baseline. 5.1 PASCAL AIRPLANE POSE ESTIMATION We begin by demonstrating the applicability of our approach on a simple airplane azimuth estimation task. The PASCAL3D+ dataset (Xiang et al., 2014) provides additional 3D annotations for 12 rigid categories in the PASCAL VOC 201”
kPAM: KeyPoint Affordances for Category-Level Robotic Manipulation (2019, background): “There exist several datasets [29, 30] annotated with aligned geometric templates, and pose estimators [21, 29] trained on these datasets can produce a category-level pose estimation.”
Learning 3D Object Categories by Looking Around Them (2017, methodology): “For VPNet trained on aligned FrC, we adjusted the produced bounding box and viewpoint annotations in the same fashion as done for adjusting the Pascal3D annotations in sec.”
SurfNet: Generating 3D Shape Surfaces Using Deep Residual Networks (2017, methodology): “We also ran our learnt networks on the airplane and car categories of the PASCAL 3D+ [38] dataset and qualitatively compare it with the method of [16].”