MegDet: A Large Mini-Batch Object Detector

Abstract

The development of object detection in the era of deep learning, from R-CNN [11], Fast/Faster R-CNN [10, 31] to recent Mask R-CNN [14] and RetinaNet [24], mainly come from novel network, new framework, or loss design. However, mini-batch size, a key factor for the training of deep neural networks, has not been well studied for object detection. In this paper, we propose a Large Mini-Batch Object Detector (MegDet) to enable the training with a large mini-batch size up to 256, so that we can effectively utilize at most 128 GPUs to significantly shorten the training time. Technically, we suggest a warmup learning rate policy and Cross-GPU Batch Normalization, which together allow us to successfully train a large mini-batch detector in much less time (e.g., from 33 hours to 4 hours), and achieve even better accuracy. The MegDet is the backbone of our submission (mmAP 52.5%) to COCO 2017 Challenge, where we won the 1st place of Detection task.

Keywords

Computer scienceNormalization (sociology)Object detectionDetectorDeep learningTask (project management)Object (grammar)Artificial intelligenceKey (lock)Training setArtificial neural networkPattern recognition (psychology)Operating systemEngineeringTelecommunications

Affiliated Institutions

Related Publications

Libra R-CNN: Towards Balanced Learning for Object Detection

Jiangmiao Pang , Kai Chen , Jianping Shi +3 more

Compared with model architectures, the training process, which is also crucial to the success of detectors, has received relatively less attention in object detection. In this w...

2019 2019 IEEE/CVF Conference on Computer ... 1634 citations

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

Sergey Ioffe , Christian Szegedy

Training Deep Neural Networks is complicated by the fact that the distribution of each layer's inputs changes during training, as the parameters of the previous layers change. T...

2024 arXiv (Cornell University) 15635 citations

Fast R-CNN

Ross Girshick

This paper proposes a Fast Region-based Convolutional Network method (Fast R-CNN) for object detection. Fast R-CNN builds on previous work to efficiently classify object proposa...

2015 2015 IEEE International Conference on... 26511 citations

Fast R-CNN

Ross Girshick

This paper proposes a Fast Region-based Convolutional Network method (Fast R-CNN) for object detection. Fast R-CNN builds on previous work to efficiently classify object proposa...

2015 arXiv (Cornell University) 1766 citations

FCOS: Fully Convolutional One-Stage Object Detection

Zhi Tian , Chunhua Shen , Hao Chen +1 more

We propose a fully convolutional one-stage object detector (FCOS) to solve object detection in a per-pixel prediction fashion, analogue to semantic segmentation. Almost all stat...

2019 5672 citations

Publication Info

Year: 2018
Type: article
Pages: 6181-6189
Citations: 316
Access: Closed

External Links

View on DOI.org

Social Impact

Altmetric

MegDet: A Large Mini-Batch Object Detector

PlumX Metrics

Social media, news, blog, policy document mentions

Citation Metrics

316

OpenAlex

Cite This

APA Style

                            
                                    Chao Peng, 
                                
                                    Tete Xiao, 
                                
                                    Zeming Li
                                
                                et al.
                            
                            (2018). 
                            MegDet: A Large Mini-Batch Object Detector. 
                            
                            , 6181-6189.
                            https://doi.org/10.1109/cvpr.2018.00647

Identifiers

DOI: 10.1109/cvpr.2018.00647