Open x-Sapphire opened 4 years ago
hello,how do you solve it? I met it too
got the same problem, do you solve it?? thank you!
hello,how do you solve it? I met it too
got the same problem, do you solve it?? thank you!
same!Help! This Only happens when using caffe pre-trained
I was able to mitigate this issue by using the Adam optimizer (setting the use_adam
bool in config to True
).
I hope this helps.
I was able to mitigate this issue by using the Adam optimizer (setting the
use_adam
bool in config toTrue
).I hope this helps.
It works! Thanks. This is a rather wired problem, I couldn't figure out why the bbox_tools is affected by optimizers.
Updates. I changed the optimizer to Adam according to jaypmorgan's suggestion. The training and loss curves seem normal now, however, I am still encountering these warnings, even though my training process is not influenced by these warnings...
got the same problem
/simple-faster-rcnn-pytorch-master/model/utils/bbox_tools.py:68: RuntimeWarning: overflow encountered in exp h = xp.exp(dh) src_height[:, xp.newaxis] /simple-faster-rcnn-pytorch-master/model/utils/bbox_tools.py:68: RuntimeWarning: overflow encountered in multiply h = xp.exp(dh) src_height[:, xp.newaxis] /simple-faster-rcnn-pytorch-master/model/utils/bbox_tools.py:69: RuntimeWarning: overflow encountered in exp w = xp.exp(dw) src_width[:, xp.newaxis] /simple-faster-rcnn-pytorch-master/model/utils/bbox_tools.py:69: RuntimeWarning: overflow encountered in multiply w = xp.exp(dw) src_width[:, xp.newaxis]
and in the visdom I cannot see any loss with cls or loc