Open m-radfar opened 8 years ago
I have the same thing, it looks like the HDF5 luarocks module is different from supported versions and Neuraltalk refuses to run because of this.
EDIT: Sorry about that, this is a torch-rnn issue: https://github.com/jcjohnson/torch-rnn/issues/105
SOLUTION for hdf5 1.10 unsupported version and GCC not compiling correctly: (For Arch-Linux (64-bit) users)
There is a package hdf5_18, which matches the last supported version of hdf5 for torch. I installed the package: yaourt hdf5_18
(coupled with yaourt h5py
)
Then updated torch/install/share/lua/5.1/hdf5/config.lua
HDF5_LIBRARIES = "/usr/lib/hdf5_18/libhdf5.so"
Now, on ~/torch/install/share/lua/5.1/hdf5/ffi.lua on line 39:
local headerPath = path.join(includePath, "hdf5_18/hdf5.h")
Change the second parameter of method join from the default "hdf5" ---> "hdf5_18/hdf5.h"
Then with CUDA (v7.5) properly installed (I've done it with yaourt aswell, it will be installed in /opt/cuda/) comes the installation of cunn and cutorch, which don't work with the latest gcc version (6.1) but do work with gcc5 (which is installed when installing the cuda toolkit) and run the commands:
CC=gcc-5 CXX=g++-5 luarocks install cutorch
CC=gcc-5 CXX=g++-5 luarocks install cunn
After this, download cudNN from NVIDIA website, extract it wherever you want but remember the path because it will be needed to add this line in ~/.bashrc:
export CUDNN_PATH=/path_to_folder_extracted/lib64/libcudnn.so.5
Proceeding:
git clone https://github.com/soumith/cudnn.torch.git
cd inside cudnn.torch and:
luarocks make cudnn-scm-1.rockspec
Finally it's only missing the: luarocks install loadcaffe
Back to the neuraltalk2 directory, run the th eval.lua with your model and images and one little repair to the README.md, SimpleHTTPServer doesn't work anymore in Python3, it was merged to http.server module so now it is (inside vis):
python -m http.server
@m-radfar hi,i have the same error with you.have you solved it yet?thanks~
after installing the: torch nn nngraph image lua-cjson-2.1.0
itried to using following command: th eval.lua -model /.../.../model_id1-501-1448236541.t7 -image_folder images/ -num_images 2
but I get an error like below:
home/.../torch/install/bin/lua: cannot open eval.lua: No such file or directory stack traceback: [C]: in function 'dofile' .../torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk [C]: in ?
I didn't find any probable requirements.