MVCNN-TensorFlow icon indicating copy to clipboard operation
MVCNN-TensorFlow copied to clipboard

Published 20 hours ago • verified publisher icon WeiTang114

Model diverged with loss = NaN

Open ttyang1018 opened this issue 7 years ago • 7 comments

Hi there,

I manually generated the 12 views of each airplane and bathtub png image, and train with them. But I encountered following model diverged problem.

image

Could someone help me out of this？

ttyang1018 avatar Jul 27 '17 15:07 ttyang1018

I have the same problem!

yifanfeng97 avatar Aug 20 '17 05:08 yifanfeng97

Hi... I wonder if this problem has been solved?? I have encountered the same problem...

foxfromworld avatar Feb 21 '18 03:02 foxfromworld

I have solved my nan problem. My reason is that my dataset is converted from MVCNN(MATLAB code). The label in MATLAB is in range 1 and 40. So I reduced 1 on label converting it into range 0 and 39.

yifanfeng97 avatar Feb 21 '18 04:02 yifanfeng97

Thank you so much for your quick response! May I ask what the labels are? Type of objects? I am sorry...I am very new to NN... Thank you again!

foxfromworld avatar Feb 21 '18 05:02 foxfromworld

yes, the label is a number to represent the name of objects in training.

yifanfeng97 avatar Feb 21 '18 06:02 yifanfeng97

Thanks a million. Really appreciate your help! I will try to change the way I feed the data and see if it works!

foxfromworld avatar Feb 21 '18 07:02 foxfromworld

I have solved my nan problem. My reason is that my dataset is converted from MVCNN(MATLAB code). The label in MATLAB is in range 1 and 40. So I reduced 1 on label converting it into range 0 and 39.

Thanks a lot. I solved this by your comment. The reason is when I processed the data, I labeled the categories from 1 to 40. Changing them back to 0 to 39 fixes this problem.

ghost avatar Jun 06 '19 02:06 ghost