search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.7k stars 7.96k forks source link

false positives in person-car detection #8517

Open wzf19947 opened 2 years ago

wzf19947 commented 2 years ago

Hello, I'm doing person-car detection while I want to detect person riding a bike or motorcycle as person with the minimum envelope box of person and bike, and I labeled my datasets like this way. But when I detected videos, it got lots of false positives happened at the front and rear of car with tyre, false detected as a person. is it because of my labeling way is wrong? Anybody could help me? Wish for any of your reply,Thanks.

akashAD98 commented 2 years ago

@wzf19947 what's the size of your data? can you please upload sample pics of your annotations & also false predictions?

wzf19947 commented 2 years ago

@akashAD98 IMG_20220517_112927 IMG_20220517_112823 IMG_20220517_113019 as you can see, I labeled person riding bike or motor as a person, but I got many false positives on car.

akashAD98 commented 2 years ago

i think your image quality is poor, & you need to work on the dataset at least you need to collect 2k instances of each class add images in your training data, which you want to actually detect. , & also you can avoid false prediction using negative sampling(baground image) adding technique,