CornerNet- Detecting Objects as Paired Keypoints

Explore an in-depth explanation of the CornerNet object detection algorithm in this 25-minute video. Dive into a novel approach that detects object bounding boxes as paired keypoints, eliminating the need for anchor boxes. Learn about the innovative corner pooling technique and its impact on localizing corners. Understand the pipeline, including the hourglass network, heatmap and embedding outputs, loss functions, and experimental results. Gain insights into how CornerNet outperforms existing one-stage detectors on the MS COCO dataset. Follow along with a detailed breakdown of the paper, complete with visual aids and explanations of key concepts in object detection and computer vision.

- Intro & High-Level Overview
- Object Detection
- Pipeline I - Hourglass
- Heatmap & Embedding Outputs
- Heatmap Loss
- Embedding Loss
- Corner Pooling
- Experiments