Bottom-up Object Detection by Grouping Extreme and Center Points

Abstract

With the advent of deep learning, object detection drifted from a bottom-up to a top-down recognition problem. State of the art algorithms enumerate a near-exhaustive list of object locations and classify each into: object or not. In this paper, we show that bottom-up approaches still perform competitively. We detect four extreme points (top-most, left-most, bottom-most, right-most) and one center point of objects using a standard keypoint estimation network. We group the five keypoints into a bounding box if they are geometrically aligned. Object detection is then a purely appearance-based keypoint estimation problem, without region classification or implicit feature learning. The proposed method performs on-par with the state-of-the-art region based detection methods, with a bounding box AP of 43.2% on COCO test-dev. In addition, our estimated extreme points directly span a coarse octagonal mask, with a COCO Mask AP of 18.9%, much better than the Mask AP of vanilla bounding boxes. Extreme point guided segmentation further improves this to 34.6% Mask AP.

References in corpus (7)

• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• YOLOv3: An Incremental Improvement
• R-FCN: Object Detection via Region-based Fully Convolutional Networks
• Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
• DSSD : Deconvolutional Single Shot Detector
• Associative Embedding: End-to-End Learning for Joint Detection and Grouping
• Point Linking Network for Object Detection

Cited by in corpus (23)

• AutoAssign: Differentiable Label Assignment for Dense Object Detection
• MIDeepSeg: Minimally Interactive Segmentation of Unseen Objects from Medical Images Using Deep Learning
• Review of data analysis in vision inspection of power lines with an in-depth discussion of deep learning technology
• RepPoints V2: Verification Meets Regression for Object Detection
• Revisiting Feature Alignment for One-stage Object Detection
• Decoupled Classification Refinement: Hard False Positive Suppression for Object Detection
• AugFPN: Improving Multi-scale Feature Learning for Object Detection
• Rethinking Classification and Localization for Object Detection
• RelationNet++: Bridging Visual Representations for Object Detection via Transformer Decoder
• Learning Human-Object Interaction Detection using Interaction Points
• CentripetalNet: Pursuing High-quality Keypoint Pairs for Object Detection
• LightTrack: A Generic Framework for Online Top-Down Human Pose Tracking
• Vertebra-Focused Landmark Detection for Scoliosis Assessment
• Monocular, One-stage, Regression of Multiple 3D People
• CenterMask: single shot instance segmentation with point representation
• Learning from Noisy Anchors for One-stage Object Detection
• X-LineNet: Detecting Aircraft in Remote Sensing Images by a pair of Intersecting Line Segments
• Oriented Objects as pairs of Middle Lines
• Decoupled IoU Regression for Object Detection
• Inter-Homines: Distance-Based Risk Estimation for Human Safety
• SAIS: Single-stage Anchor-free Instance Segmentation
• Compact Scene Graphs for Layout Composition and Patch Retrieval
• Perceiving Traffic from Aerial Images