The install process assumes that CUDA and CUDNN are installed.

The build_env script assumes you are using Ubuntu. If you are not, you are on your own.

The script build_env builds the environment for running the demo This should be turned into a container someday. If you have the following dependencies satisified, you do not have to run the build_env script

• nvcaffe (at least 0.15.x)

• opencv (at least 2.4.11 with ffmpeg support built in)

• the various python packages referenced in requirements.txt

• the various packages referenced in build_env

• Creating the environment

  virtualenv venv

  source venv/bin/activate

  (venv) # sh build_env

Running the Demo:

(venv) # ./demo_video_infer.py

Leaving the environment

(venv) # deactivate

PATH TO SET:

export PYTHONPATH=pwd/venv/python/2.7:pwd/caffe/python export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

Generate Video:

ffmpeg -r 30 -i 'outdata/outfile.%05d.jpg' -c:v libx264 -qp 0 full_inference_v3.mp4