Detecting Objects on Multiple Images using You Only Look Once (Yolo)



When I started to perform my fourth studies in 2018, i.e. counting people in the Instagram images, I found the popular object detection program Yolo, which is short for a cool name “You only look once”. I was excited by this name and also at the moment when I made it running. However, I found either the original version or the widely used spin-off (AlexeyAB) of the Yolo only support object detection from one image per time. This leads to incredible inconvenience in my study, as I need to detect people in thousands of social media images.

After exploring this problem in the Yolo community, I realised I’m not the only one with this problem. Detecting objects in multiple images is necessary but not yet implemented in Yolo.

At the end of that day, I decided to do something for this community, i.e. improving Yolo by adding the batch image detecting function. Immediately I started reading code in the next sunny afternoon and then designing functions. The first released version supports detecting objects in multiple images located in a folder. According to the requirements from the community, I continue adding new features. In the most recent releasing, it supports exporting information of detected objects, such as name and bounding box, in either text and JSON format.

For more detail, please visit the website for this project. Like it? Give it a star on Github.


It was fun for me when doing this project. Since I left my first job ten years ago, I haven’t performed any C programming. Though I’m not bad with programming in C, I don’t like it that much. I can bear with it but won’t do it as a job. I can still remember the moment I quit my first job (team) I thought that was the last day in my life to programming in C. Too naive obviously, I did not expect after ten years I still have to do some C programming, though it is not that intensive. I was even writing C with music and recalled all the funny stories from my first job. It was nice.

comments powered by Disqus