Fast Single Shot Detection and Pose Estimation

By Patrick Poirson, Phil Ammirato, Cheng-Yang Fu, Jana Kosecka, Alexander C. Berg. 2016

112 citations recorded by the source corpus.

For applications in navigation and robotics, estimating the 3D pose of objects is as important as detection. Many approaches to pose estimation rely on detecting or tracking parts or keypoints [11, 21]. In this paper we build on a recent state-of-the-art convolutional network for sliding-window detection [10] to provide detection and rough pose estimation in a single shot, without intermediate stages of detecting parts or initial bounding boxes. While not the first system to treat pose estimation as a categorization problem, this is the first attempt to combine detection and pose estimation at the same level using a deep learning approach. The key to the architecture is a deep convolutional network where scores for the presence of an object category, the offset for its location, and the approximate pose are all estimated on a regular grid of locations in the image. The resulting system is as accurate as recent work on pose estimation (42.4% 8 View mAVP on Pascal 3D+ [21] ) and significantly faster (46 frames per second (FPS) on a TITAN X GPU). This approach to detection and rough pose estimation is fast and accurate enough to be widely applied as a pre-processing step for tasks including high-accuracy pose estimation, object tracking and localization, and vSLAM.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

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

Citations in the corpus

Speed/Accuracy Trade-Offs for Modern Convolutional Object Detectors (2016, background): “, [27] extended this idea to predict boxes, classes and pose.”
Lift, Splat, Shoot: Encoding Images From Arbitrary Camera Rigs by Implicitly Unprojecting to 3D (2020, methodology): “A standard technique is to apply a mature 2D object detector in the image plane and then train a second network to regress 2D boxes into 3D boxes [12,26,31,27].”
Skeleton boxes: Solving skeleton based action detection with a single deep convolutional neural network (2017, methodology): “Our work is also related to the SSD [9] which is an effective general image object detection method, and widely utilized in other image detection work [12, 13].”
SSD-6D: Making RGB-Based 3D Detection and 6D Pose Estimation Great Again (2017, methodology): “As for works regressing the pose from RGB images, the related works of [24, 23] recently extended SSD to include pose estimates for categories.”
Real-Time Seamless Single Shot 6D Object Pose Prediction (2017, background): “SSD has been extended to predict the object’s identity, its 2D bounding box in the image and a discrete estimate of the object’s orientation [10, 24].”
3D Bounding Box Estimation Using Deep Learning and Geometry (2016, methodology): “Given the coarse estimate of the pose obtained from a DPMbased detection pipeline, the continuous 6 DoF pose is refined by estimating the correspondences between the projected 3D model and the image contours.”
A dataset for developing and benchmarking active vision (2017, background): “The state of the art for object classification and recognition in these datasets is based on either object proposals and feature pooling following [7] with advanced deep networks [8], [9] or on fully convolutional networks implementing a modern take on sliding windows [3], [10], [11] that provide fr”
A Survey of Deep Learning: Platforms, Applications and Emerging Research Trends (2018, background): “[105] developed a model for simultaneous 3D object detection and pose estimation in a single deep CNN, greatly increasing the efficiency of detection and pose estimation over other state-of-the-art systems.”
Orthographic Feature Transform for Monocular 3D Object Detection (2018, methodology): “Many approaches start from 2D bounding boxes extracted using standard detectors described above, upon which they either directly regress 3D pose parameters for each region [14, 26, 24, 23] or ﬁt 3D templates to the image [2, 35, 36, 38].”
3D-RCNN: Instance-Level 3D Object Reconstruction via Render-and-Compare (2018, background): “Many recent works have addressed instance-level 3D scene understanding [36, 4, 6, 29, 38, 33, 50, 47, 56, 64, 35, 2, 63, 62, 12].”