search

zhaoweicai / mscnn

Caffe implementation of our multi-scale object detection framework
405 stars 211 forks source link

Reason why you didn't train cars and pedestrian together? #25

Closed jinuhjoo closed 7 years ago

jinuhjoo commented 8 years ago

Hello @zhaoweicai I was wondering if there was a reason why you didn't train cars and pedestrian together, but trained them separately?

Thanks and have a great day~

zhaoweicai commented 8 years ago

Hi @jinuhjoo I was following some other papers. There is no constraint to train a single model. But there are two issues. 1. The numbers of training samples are seriously imbalanced. Car has many more annotations than pedestrian and cyclist. 2. The evaluation metrics are different for car (IOU>=0.7) and ped/cyc (IOU>=0.5). You can give it a try to see how it works.

jinuhjoo commented 8 years ago

@zhaoweicai OK thanks, already on it ~~ hope it will give reasonable results~~