Acquisition of Localization Confidence for Accurate Object Detection
arXiv:1807.11590
Abstract
Modern CNN-based object detectors rely on bounding box regression and non-maximum suppression to localize objects. While the probabilities for class labels naturally reflect classification confidence, localization confidence is absent. This makes properly localized bounding boxes degenerate during iterative regression or even suppressed during NMS. In the paper we propose IoU-Net learning to predict the IoU between each detected bounding box and the matched ground-truth. The network acquires this confidence of localization, which improves the NMS procedure by preserving accurately localized bounding boxes. Furthermore, an optimization-based bounding box refinement method is proposed, where the predicted IoU is formulated as the objective. Extensive experiments on the MS-COCO dataset show the effectiveness of IoU-Net, as well as its compatibility with and adaptivity to several state-of-the-art object detectors.
arXiv commentAccepted to European Conference on Computer Vision (ECCV) 2018
References in corpus (6)
- DeepPose: Human Pose Estimation via Deep Neural Networks
- Learning to Segment Object Candidates
- DenseBox: Unifying Landmark Localization with End to End Object Detection
- Cascade R-CNN: Delving into High Quality Object Detection
- Relation Networks for Object Detection
- Repulsion Loss: Detecting Pedestrians in a Crowd
Cited by in corpus (17)
- Libra R-CNN: Towards Balanced Learning for Object Detection
- Center-based 3D Object Detection and Tracking
- Decoupled Classification Refinement: Hard False Positive Suppression for Object Detection
- PyramidBox++: High Performance Detector for Finding Tiny Face
- GS3D: An Efficient 3D Object Detection Framework for Autonomous Driving
- PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation
- CentripetalNet: Pursuing High-quality Keypoint Pairs for Object Detection
- FoodLogoDet-1500: A Dataset for Large-Scale Food Logo Detection via Multi-Scale Feature Decoupling Network
- Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection
- Solution for Large-Scale Hierarchical Object Detection Datasets with Incomplete Annotation and Data Imbalance
- Objects in Semantic Topology
- Confluence: A Robust Non-IoU Alternative to Non-Maxima Suppression in Object Detection
- Localization Distillation for Dense Object Detection
- High-Performance Long-Term Tracking with Meta-Updater
- Learning Global Structure Consistency for Robust Object Tracking
- Decoupled IoU Regression for Object Detection
- Saliency-Associated Object Tracking