search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.8k stars 7.97k forks source link

mAP 0 for certain classes #7620

Open kartik1395 opened 3 years ago

kartik1395 commented 3 years ago

I’m training a model with 18 classes. I used intermediate weights to test the model to see how it is performing. But mAP of a lot of classes is 0 even though images are present in the test set. 9E25B089-68C2-4666-B784-3FDF5B6AAA83

poornimajd commented 3 years ago

Do you have these classes(whose mAP is 0) in good number in the test images?

kartik1395 commented 3 years ago

Yup i do.

Regards Kartik Chopra

On 20 Apr 2021, at 17:42, Poornima JD @.***> wrote:



Do you have these classes(whose mAP is 0) in good number in the test images?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/AlexeyAB/darknet/issues/7620#issuecomment-823136741, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AMQBFHJ6MONUYKVBZCTQAZDTJVD6NANCNFSM43HYJWFQ.

DISCLAIMER The sender of this email is an alumnus of National University of Singapore (NUS). Kindly note that NUS is not responsible for the contents of this email, and views and opinions expressed are solely the sender's.

poornimajd commented 3 years ago

I think you can try to overfit the model for a small number of images of your dataset i.e use the same data for test and train and see if the model is really learning these classes.just to rule out some other error possibilities.

kartik1395 commented 3 years ago

Thank you for the input. I realised that the annotations were overlapping. After class 9 it started from 0 again.

Regards Kartik Chopra

On 20 Apr 2021, at 17:48, Poornima JD @.***> wrote:



I think you can try to overfit the model for a small number of images of your dataset i.e use the same data for test and train and see if the model is really learning these classes.just to rule out some other error possibilities.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/AlexeyAB/darknet/issues/7620#issuecomment-823141089, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AMQBFHOX5EO32Z6WBVSYW5LTJVEYJANCNFSM43HYJWFQ.

DISCLAIMER The sender of this email is an alumnus of National University of Singapore (NUS). Kindly note that NUS is not responsible for the contents of this email, and views and opinions expressed are solely the sender's.

sunzhoujun commented 3 years ago

I encountered a similar problem with you. How did you solve it？