Omni3D: A Large Benchmark and Model for 3D Object Detection in the Wild

by   Garrick Brazil, et al.

Recognizing scenes and objects in 3D from a single image is a longstanding goal of computer vision with applications in robotics and AR/VR. For 2D recognition, large datasets and scalable solutions have led to unprecedented advances. In 3D, existing benchmarks are small in size and approaches specialize in few object categories and specific domains, e.g. urban driving scenes. Motivated by the success of 2D recognition, we revisit the task of 3D object detection by introducing a large benchmark, called Omni3D. Omni3D re-purposes and combines existing datasets resulting in 234k images annotated with more than 3 million instances and 97 categories.3D detection at such scale is challenging due to variations in camera intrinsics and the rich diversity of scene and object types. We propose a model, called Cube R-CNN, designed to generalize across camera and scene types with a unified approach. We show that Cube R-CNN outperforms prior works on the larger Omni3D and existing benchmarks. Finally, we prove that Omni3D is a powerful dataset for 3D object recognition, show that it improves single-dataset performance and can accelerate learning on new smaller datasets via pre-training.


page 1

page 7

page 13

page 14


Peng Cheng Object Detection Benchmark for Smart City

Object detection is an algorithm that recognizes and locates the objects...

BigDetection: A Large-scale Benchmark for Improved Object Detector Pre-training

Multiple datasets and open challenges for object detection have been int...

Towards Large-Scale Small Object Detection: Survey and Benchmarks

With the rise of deep convolutional neural networks, object detection ha...

The EuroCity Persons Dataset: A Novel Benchmark for Object Detection

Big data has had a great share in the success of deep learning in comput...

IIIT-AR-13K: A New Dataset for Graphical Object Detection in Documents

We introduce a new dataset for graphical object detection in business do...

The Open Brands Dataset: Unified brand detection and recognition at scale

Intellectual property protection(IPP) have received more and more attent...

An optical biomimetic eyes with interested object imaging

We presented an optical system to perform imaging interested objects in ...

Please sign up or login with your details

Forgot password? Click here to reset