Rethinking Classification and Localization for Object Detection

By Yue Wu, Yinpeng Chen, Lu Yuan, Zicheng Liu, Lijuan Wang, Hongzhi Li, Yun Raymond Fu. Computer Vision and Pattern Recognition, 2019

620 citations recorded by the source corpus.

Two head structures (i.e. fully connected head and convolution head) have been widely used in R-CNN based detectors for classification and localization tasks. However, there is a lack of understanding of how does these two head structures work for these two tasks. To address this issue, we perform a thorough analysis and find an interesting fact that the two head structures have opposite preferences towards the two tasks. Specifically, the fully connected head (fc-head) is more suitable for the classification task, while the convolution head (conv-head) is more suitable for the localization task. Furthermore, we examine the output feature maps of both heads and find that fc-head has more spatial sensitivity than conv-head. Thus, fc-head has more capability to distinguish a complete object from part of an object, but is not robust to regress the whole object. Based upon these findings, we propose a Double-Head method, which has a fully connected head focusing on classification and a convolution head for bounding box regression. Without bells and whistles, our method gains +3.5 and +2.8 AP on MS COCO dataset from Feature Pyramid Network (FPN) baselines with ResNet-50 and ResNet-101 backbones, respectively.

Subfields: Artificial Intelligence, Biomedical Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/3c926f0f307cf24259ae3a9721736bee00d74c63

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Lijuan Wang
2. Jianfeng Gao
3. Zicheng Liu
4. Jianfeng Wang
5. Xiaowei Hu
6. Pengchuan Zhang
7. Zhe Gan
8. Jianwei Yang

In Biomedical Engineering

1. Liang Wang
2. Zhang Zhang
3. Tien-Ping Tan
4. Zicheng Liu
5. Gong Cheng
6. Junwei Han
7. Wanqing Li
8. Hossein Rahmani

In Computer Vision and Pattern Recognition

1. Jianfeng Gao
2. Dacheng Tao
3. Zicheng Liu
4. Lijuan Wang
5. Junwei Han
6. Luc Van Gool
7. Jianfeng Wang
8. Gong Cheng

Citations in the corpus

• MiniVLM: A Smaller and Faster Vision-Language Model (2020, background): “Region feature is built on top of the object detector, and the detector can be two-stage [38, 8, 24, 55] or one-stage [37, 48, 29, 27, 6, 58, 53, 49].”
• A Comprehensive Review of YOLO: From YOLOv1 to YOLOv8 and Beyond (2023, background): “In [82, 83], it was shown that there could be a misalignment between the classification confidence and localization accuracy.”
• UP-DETR: Unsupervised Pre-training for Object Detection with Transformers (2020, background): “Object detection is the coupling of object classification and localization, where these two tasks always have different feature preferences [45, 40, 35].”
• Anchor-Free Oriented Proposal Generator for Object Detection (2021, methodology): “To further boost the performance and evaluate the consistency of the effectiveness of AOPG, we supplement the experiments based on more advanced frameworks like Cascade R-CNN [17] and Double Head [54].”
• Align Deep Features for Oriented Object Detection (2020, background): “Double-Head R-CNN [29] adopts different head architectures for different tasks, i.”
• DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution (2020, neutral): “, [5,7,8,10,26,27, 29,34,62,75].”
• QueryDet: Cascaded Sparse Query for Accelerating High-Resolution Small Object Detection (2021, neutral): “3 Double-Head-Ext [57] CVPR 2020 ResNet-101 42.”
• Focal and Efficient IOU Loss for Accurate Bounding Box Regression (2021, methodology): “Till now, BBR has been used on most of the recent detection methods [5, 8, 21, 24, 1, 16, 6].”
• DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection (2021, background): “2 may lead to possible suboptimal solution on individual tasks in order to balance the mismatched goals of them [7, 49].”