Posts /

YOLO and its evolution

Twitter Facebook
02 Jan 2018


YOLO(You Only Look Once)

1. Introduction

Prior work on object detection repurpose classifiers to perform detection. 即基于 DPM 或者 RPN 的检测方法,这些方法首先生成一系列供选择的 region proposals,然后在这些 RP 上运行一个 object classifier,最后还需要对预测出的 bounding boxes 进行 refine。这种方法相比 YOLO 可以达到更高的准确率,不过它把整个 object detection 的 pipeline 分割成了几部分,这导致这种方法速度很慢,而且难以优化,因为三个部分需要分别进行训练。

YOLO reframes object detection as a single regression problem. Straight from image pixels to bounding box coordinates and class probabilities. This unified model has several benefits over traditional methods of object detection.

For a demo of YOLO in real-time on a web cam, you can visit the project webpage: YOLO PROJECT.

2. Unified Detection

2.1 Network Design


Twitter Facebook