3DNet: Large-scale object class recognition from CAD models

By Walter Wohlkinger, Aitor Aldoma, Radu Bogdan Rusu, Markus Vincze. 2012

178 citations recorded by the source corpus.

Subfields: Aerospace Engineering, 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 Aerospace Engineering

In Computer Vision and Pattern Recognition

In Geology

Citations in the corpus

RGB-D datasets using microsoft kinect or similar sensors: a survey (2016, methodology): “All images are recorded through a single Kinect sensor which is mounted on the top of a Willow Garage PR2 (robot).”
Dex-Net 2.0: Deep Learning to Plan Robust Grasps with Synthetic Point Clouds and Analytic Grasp Metrics (2017, neutral): “0: 1,371 synthetic models from 3DNet [63] and 129 laser scans from the KIT object database [26].”
Learning to Detect Good 3D Keypoints (2017, background): “…of comparative evaluations addressing object recognition (Alexandre 2012; Aldoma et al. 2012a), 3D object classification and retrieval (Proença et al. 2013; Wohlkinger et al. 2012), semantic segmentation of point clouds (Behley et al. 2012), localization ofmedical landmarks (Sukno et al. 2012).”
Hobbit, a care robot supporting independent living at home: First prototype and lessons learned (2016, methodology): “15(b)) and while rotating the arm each new view of the object is stored in a database [41] and later matched against in the recognition phase using random forests [42].”
Learning a visuomotor controller for real world robotic grasping using simulated depth images (2017, methodology): “The objects were derived from CAD models contained within the 3DNet database [16].”
Leveraging big data for grasp planning (2015, methodology): “Most of the mesh models are from the 3DNet database [38] with further augmentations from [1] [5].”
Grasp Pose Detection in Point Clouds (2017, methodology): “Compared with the amount of real sensor data that is available, there is a huge number of CAD models available online (e.g., 3DNET makes available thousands of CAD models from 200 object categories (Wohlkinger et al. (2012)).”
High precision grasp pose detection in dense clutter (2016, methodology): “This raw data can be obtained using online CAD models such as those contained in 3DNET [19] and simulating the depth images, or it can come from a dataset like BigBird that pairs real depth images with a reconstructed mesh of the object.”
Dex-Net 3.0: Computing Robust Vacuum Suction Grasp Targets in Point Clouds Using a New Analytic Model and Deep Learning (2018, methodology): “Our GQ-CNN trained on synthetic data from 3DNet [33], KIT [15], and the Adversarial objects.”
Dex-Net 1.0: A cloud-based network of 3D objects for robust grasp planning using a Multi-Armed Bandit model with correlated rewards (2016, methodology): “0 contains 13,252 3D mesh models: 8,987 from the SHREC 2014 challenge dataset [28], 2,539 from ModelNet40 [45], 1,371 from 3DNet [44], 129 from the KIT object database∗ [19], 120 from BigBIRD∗ [38], 80 from the Yale-CMU-Berkeley dataset∗ [8], and 26 from the Amazon”
Learning Dexterous Grasping with Object-Centric Visual Affordances (2020, methodology): “Grasping unseen objects from 3DNet Next we push the robustness challenge further by requiring the agent to generalize its grasp behavior to objects it has not encountered before (the 24 3DNet [52] objects).”
Pick and Place Without Geometric Object Models (2017, methodology): “To do so, we obtained objects belonging to two different categories for experimentation: a set of 73 bottles and a set of 75 mugs – both in the form of mesh models from 3DNet [30].”
Dex-Net 3.0: Computing Robust Robot Suction Grasp Targets in Point Clouds using a New Analytic Model and Deep Learning (2017, methodology): “A GQ-CNN trained on synthetic data from 3DNet [35], KIT [16], and the Adversarial objects.”
SHOT: Unique signatures of histograms for surface and texture description (2014, methodology): “Finally, SHOT has been deployed also as a global descriptor in the context of 3D object retrieval from a large 3D CAD database [55]: in the proposed comparison to other global descriptors it turned out the second-best approach in terms of retrieval accuracy.”