Problem about loss staying in NAN

[Cheren15]

Hi, I came across the case that x_hat, h_hat, w_hat and loss are staying at NAN at each step, I followed some issue and deleted problematic file 0132 and 0165(all 4 files in raw dataset) and then create the tf-record again. Unfortunately, I still got the same problem. BTW, I am using python 2.7 and tensorflow 1.6.

[Cheren15]

And I have directly used the grasp weights train on grasp dataset, can anybody help me out, thank you!