AlexeyAB
commented
6 years ago Hi,
- The best way if you train your own model for one class (person) on your image-dataset that got from the same camera which you use to detect
- Also you can use already trained models VOC or COCO models. Don't use Yolo9000 or tiny models for this.
(if you want to detect only person without any other objects, then you can change 2 lines in the file
image.cand recompile, as described here: https://github.com/AlexeyAB/darknet/issues/229#issuecomment-336309483 )
Hi! Thank you for the work you put down into this.
I am using this to run object identification on the videos that my surveillance cameras are recording and if a person is detected i get a notification on my phone. I just wanted to contact you and ask if I am running in an optimized way for only finding persons in recording?
There are many options that you describe in your readme which makes me wonder what way is best if i only am interested in finding persons in the video. VOC, YOLO9000 or COCO model? should i use tiny YOLO?
Today i run like this in my python script:
["./darknet", "detector", "demo", "cfg/coco.data", "cfg/yolo.cfg", "yolo.weights", filepath, "-i", "0", "-out_filename", "result.avi"]
Thank you for your time.