Error in output on Jetson

nitish11 commented 8 years ago

Context of the issue.

I am setting up a sample example environment on NVIDIA Jetson tegra.
I am not getting any error in building. I am getting error in output. The representations coming out from deep net trained model are not as expected.

Command executed :

root@tegra-ubuntu:/home/ubuntu/face-recognition/openface# ./demos/classifier.py infer ./models/openface/celeb-classifier.nn4.small2.v1.pkl images/examples/{carell,adams,lennon}*

Actual behavior.

=== images/examples/carell.jpg ===
Predict BenStiller with 0.67 confidence.

=== images/examples/adams.jpg ===
Predict BenStiller with 0.67 confidence.

=== images/examples/lennon-1.jpg ===
Predict BenStiller with 0.67 confidence.

=== images/examples/lennon-2.jpg ===
Predict BenStiller with 0.67 confidence.

OS and hardware information.

Operating system: Ubuntu 14.04
Torch version: torch7
CPU architecture: 64-bit
Device : Linux tegra-ubuntu 3.10.96-tegra aarch64 GNU/Linux

bamos commented 8 years ago

This is potentially related to https://groups.google.com/forum/#!topic/cmu-openface/O029A5QQDm0. Torch temporarily had an issue that caused the network to always output the same value regardless of the input that has been fixed in the latest version. Can you re-install Torch or the nn/cunn packages?

luarocks install nn
# If using CUDA
luarocks install cunn

nitish11 commented 8 years ago

@bamos :+1: thanks. The issue was in latest torch build. Follow the below commands.

git clone https://github.com/torch/distro.git ~/torch
cd ~/torch 
git checkout f95379d8512ab96939b0d797c489bbaedf7fa313 
git submodule update --init --recursive 
./install-deps 
./install.sh 
. ~/.bashrc
luarocks install dpnn

Check Torch 7 Scientific Computer Framework with cuDNN. http://jetsonhacks.com/2015/05/20/torch-7-scientific-computer-framework-with-cudnn-nvidia-jetson-tk1/

cmusatyalab / openface