search

pjreddie / darknet

Convolutional Neural Networks
http://pjreddie.com/darknet/
Other
25.7k stars 21.33k forks source link

Partial Detection #2523

Open iamrajee opened 2 years ago

iamrajee commented 2 years ago

I'm using a YOLOv3 custom network to detect a single drone class. But after training the data I only get only a few of the drones detected. Here is the picture of the same: image

However when it works well on the fewer number of drones (<10) as below: image

What could be the reason for this anomaly? Any help would be of great help. Thank you!

Kjelldor commented 2 years ago

While I am not sure it explains the size of the disparity, part of the reason that you probably won't get all of the drones in the first image is because, unlike the second image, some drones overlap each other. Unless you train the model to detect overlapping drones as well (e.g. put examples of overlapping drones in your training data annotating them both), that will probably not improve.