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

24 Mar 2022  ยท  Likun Cai, Zhi Zhang, Yi Zhu, Li Zhang, Mu Li, xiangyang xue ยท

Multiple datasets and open challenges for object detection have been introduced in recent years. To build more general and powerful object detection systems, in this paper, we construct a new large-scale benchmark termed BigDetection. Our goal is to simply leverage the training data from existing datasets (LVIS, OpenImages and Object365) with carefully designed principles, and curate a larger dataset for improved detector pre-training. Specifically, we generate a new taxonomy which unifies the heterogeneous label spaces from different sources. Our BigDetection dataset has 600 object categories and contains over 3.4M training images with 36M bounding boxes. It is much larger in multiple dimensions than previous benchmarks, which offers both opportunities and challenges. Extensive experiments demonstrate its validity as a new benchmark for evaluating different object detection methods, and its effectiveness as a pre-training dataset.

PDF Abstract

Datasets


Introduced in the Paper:

BigDetection

Used in the Paper:

MS COCO LVIS Objects365

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Object Detection BigDetection val Cascade R-CNN (R50-FPN) AP 24.1 # 1
AP50 33.0 # 1
AP75 25.8 # 1
Object Detection BigDetection val CenterNet2 (R50-FPN) AP 23.1 # 2
AP50 30.2 # 2
AP75 24.9 # 2
Object Detection BigDetection val Faster R-CNN (R50-FPN) AP 19.4 # 3
AP50 29.3 # 3
AP75 21.3 # 3
Object Detection BigDetection val Faster R-CNN (R50) AP 18.9 # 4
AP50 28.8 # 4
AP75 20.5 # 4
Object Detection BigDetection val Deformable DETR (R50) AP 13.1 # 5
AP50 19.3 # 5
AP75 14.2 # 5
Object Detection BigDetection val YOLOv3 (D53) AP 9.7 # 6
AP50 17.4 # 6
AP75 9.7 # 6

Methods


No methods listed for this paper. Add relevant methods here