Train with FAT data

[hsisfree]

Hi, I'm trying to use fat dataset instead creating my own data.
I also changed utils_dope.py to load projected_cuboid and projected_cuboid_centroid.
I'm using train.py at train2 but I got error messages "NaN or Inf found in input tensor."

I tried to find where this error came out.
I put "print(checkpoint0)"~"print(checkpoint8)" between the code that write.add_image() at train.py.

I checked tensorboard. The images(train_Belief_ground_truth, train_Belief_guess, train_input) are stored 7 times. I trained 10 epoch. Stored 1,2,3,4,5,7,10 step.

I check steps but the number of steps are not constant....
Can I ignore this error if I will not use tensorboard?? If I have to edit the code can you give me any help please?
Additionally can I erase this part(write.add_image) if I don't need to use tensorboard??

[TontonTremblay]

I never understood how this error, and yes you can ignore it.

[hsisfree]

Thanks for your answer. I think this error came out from the code that store images at tensorboard. I checked that NaN and Inf is in "grid". But I tried with data made by nvisii, I also find NaN and Inf is grid too. I don't know why this errors came out with Fat data.

[TontonTremblay]

that would make sense. Thank you, anyway, it wont impact your network.