Open todeletel opened 2 years ago
There are more than one fish float and different types of landscapes in wow. So the accuracy will increase if you train with more screenshots. I trained a model with 200+ screenshots and got over 85% for most cases. Actually, I can get almost 100% for some particular spots. So you can either increase the training dataset or try to find some spots that have fewer distractions around.
Hi, thanks so much for providing an approach instead of cv detection. I trained a model based on yolov5s and found out my model can detect float well but just have 50%-60% to predict caught event. Now I use audio event to detect caught that work well in a silent environment only. I want to know what is the best caught rate and if I can improve the accuracy by annotating more data?