Dear Research Community.
I am new to the field of deep learning based object detection techniques. At this time, I am working on a project which will not use front-facing camera to detect people. Nearly 99% of the datasets that I have across uses some form of front-facing camera image to create its database. For the literature survey I have done, it is suggested that the YOLO model should be given as much relevant data as possible as for DNN based techniques, their accuracy scales with the number of training samples provided.
Thus, if you would suggest some tools and techniques using which I can use to create my own custom dataset, I will appreciate it.
With best regards,
Azmyin
Hello Azmyin,
Please check this article.
Title: How to Create a New Custom Dataset from Images
Url: https://medium.com/towards-artificial-intelligence/how-to-create-a-new-custom-dataset-from-images-9b95977964ab
If you want to do annotations by hand, you can use tools such as labelImg:
https://github.com/tzutalin/labelImg
This tool allows you to set the annotation mode to the YOLO style.
Once you have this, then you will need to compute the anchors using the YOLO tools, this will depend on the version of YOLO used*
Best,
Thank you Dr. Iqbal and Mr. Tobias for your valuable feedback. I will definitely check those out and post my experiences with these tools as soon as I perform the baseline test.
With best,
Azmyin
- Badges
- Science topic