everyone. I'm new to deep learning. My task is to decide if a soccer ball is inside an image(images are extracted from videos), just give true or false.
- In this case, is YOLO the best choice to solve this problem? I do not need bounding boxes. And the class number is 1, only soccer ball. So it is a two-class classification problem(contains a ball or not).
- If I use YOLO, when training, do I need to train images which do not contain a ball(thus no object)?
- What is a reasonable data size? I feel that 500000 is just too big a number.
- What is the best way to annotate? I have thousands of images(in fact, 500000), it is almost impossible to annotate by hand. Is there some automatic annotating tools?
- English is not my first language. I want to find similar projects and learn. But my description of the task is not good, and cannot get proper answers. Could you please provide me with a more precise description so that I can find similar projects?
It will be great if you can tell me what I can read to solve my questions. Thanks.
CodePudding user response:
In my experience, I would say you can use YOLO for this case. I've made a similar project based on detecting a basketball in a video frame.
If you use YOLO, you dont need images with no ball on it. I would recommend to have all kinds of pictures like pictures where the ball is moving (blurry), or in the background will make your model more robust.
If you use YOLO, you need to label your pictures (I used labelimg for this: https://github.com/tzutalin/labelImg) this can't easily be automated, but you can look for online prelabelled datasets or hire someone to do it for you. (For my own project I labelled 2000 pictures by hand and my model was working fairly well.
I did some searching and found this. (https://dbloisi.github.io/tutorial/balldetection.html) It might be helpful for you. This is not with YOLO but is also a valid option for your project.
I hope my answer has some use for you and good luck with your task!