[deep learning target detection series - 01] what is target detection

object detection (Object Detection) It has always been a hot topic in the field of deep learning , This series of articles will comb it in detail .

1. What is target detection

• The answer to the image classification task is " What is the object in the image ". The answer of the target detection task is " What is the object in the image , Where is it in the image ".
• namely , Not only output the category to which the target belongs , Also output the position information of the object . Location information usually adopts bounding box Express , That is, given the center coordinates and width and height information of a rectangular box to locate the position of the box in the image .

2. How to detect targets

• There are generally two kinds of classification from the process , namely two-stage Methods and ont-shot Methods .
• two-stage Methods , That is, to generate candidate regions first , Then classify each candidate region , Then through the return correction bounding box, Finally, the final detection result is determined by non maximum suppression .
• one-shot The main representative of the method is YOLO series , Its idea is to divide the picture into cells , Then process each cell , Directly predict whether there is an object in the current position 、bounding box The location of and the probability of classification .

The blogger will continue to update some knowledge and practice in the field of artificial intelligence 、 Problems and insights encountered in work 、 Methods and skills to work efficiently , Please pay attention if you like 、 give the thumbs-up 、 Collection support ~