search

theAIGuysCode / yolov4-deepsort

Object tracking implemented with YOLOv4, DeepSort, and TensorFlow.
GNU General Public License v3.0
1.33k stars 750 forks source link

Not detection with custom weights #112

Open jangelRiveaux opened 3 years ago

jangelRiveaux commented 3 years ago

I try to use the project with the yolov4 model of darknet project (https://github.com/AlexeyAB/darknet) trained for my own set of objects. When I used the coco weitghs of your link the traking and recognitions is ok but when I use my own weights saved with the save_model.py that you provide the algorithm return only one valid detection in the first frame, and after that it does not recognized any valid detection. Also, using your own code, I save each frame as jpeg and test the weights of the network in darknet with that images and it return succefully, so I do not understand what can be happen. I will be gratful for your help.

LukeRl commented 2 years ago

Hey, did you ever figure this out? I'm tracking people as well as a custom class and the 'person' class does fine, but the custom class is never detected. When I run the model in just darknet and test it it does detect the custom class with 90% accuracy.