Error when running eval.lua ("attempt to index field 'tensorOutput' (a nil value)") about neuraltalk2 HOT 12 OPEN

FWIW, I'm seeing exactly the same issue on Ubuntu 14.04 running in cpu-only mode with the provided CPU checkpoint model as well.

from neuraltalk2.

I have the same problem with the CPU model. Unfortunately I can't use the GPU one :(

from neuraltalk2.

Same problem here, in Vagrant

from neuraltalk2.

I don't see what could be the issue here. @soumith any ideas? I can't reproduce this error - my CPU checkpoints run fine on a Ubuntu setup. It seems very odd that tensorOutput should be nil, when it is initialized right in the constructor of nn.Identity, as seen here

edit: could this be due to some nonrecent version of torch/nn/nngraph, or some version conflict among them?

from neuraltalk2.

I followed all your build instructions on a clean Ubuntu trusty64 VM. I used this command:

th eval.lua -model ../model_id1-501-1448236541.t7_cpu.t7 -image_folder ../images -num_images 1 -gpuid -1

from neuraltalk2.

same error output as herenow, on the same setup (except I have an 2.2 GHz i7 MBP late 2011)

fresh install from torch / nn / nngraph / image

I don't have an Nvidia, so I guess in don't need/can't use CUDA / cutorch / cunn / cudnn and I need to use my cpu. Am I wrong?

from neuraltalk2.

I think I identified the issue - it's due to a Torch nn change from 8 days ago. Working out a fix.

from neuraltalk2.

hi @herenow and @aroomthedoomed

luarocks install nn

in your terminal, willl fix this issue. Sorry about that.

from neuraltalk2.

reinstalling the nn package fixed the problem for me, thanks

from neuraltalk2.

reinstalling nn fixed it for me too
thanks @soumith @karpathy

from neuraltalk2.

Had the same issue on Mac OS X 10.10.5, reinstalling nn also fixed it for me. Thanks.

from neuraltalk2.

@soumith @karpathy 👍 Fixed

from neuraltalk2.