First, I tried to train from scratch, I got NaNs after few iterations while using momentum optimizer (I also tried different learning rate) but while using adam optimizer my model is still training without NaNs. However, I am not getting correct predictions.
So, now, I am trying to start training using Imagenet weights for the Tiny Yolo v2, I got the pretrained weights from here https://pjreddie.com/darknet/imagenet/#reference . Can you please suggest me what would be the ideal training setting according to you for learning rate and optimizer to train properly?
Also, how did you get pre-trained Imagenet weights for the models you have uploaded here? Did you extract it from darknet or you trained them by yourself?
Hello @leeyoshinari
I am trying to train Tiny Yolo v2 model on VOC dataset (https://github.com/pjreddie/darknet/blob/master/cfg/yolov2-tiny.cfg) with your training procedure.
First, I tried to train from scratch, I got NaNs after few iterations while using momentum optimizer (I also tried different learning rate) but while using adam optimizer my model is still training without NaNs. However, I am not getting correct predictions.
So, now, I am trying to start training using Imagenet weights for the Tiny Yolo v2, I got the pretrained weights from here https://pjreddie.com/darknet/imagenet/#reference . Can you please suggest me what would be the ideal training setting according to you for learning rate and optimizer to train properly?
Also, how did you get pre-trained Imagenet weights for the models you have uploaded here? Did you extract it from darknet or you trained them by yourself?
Thank you. Best, Saurabh