CogStream Video Analytics

Requirements: Python3.5+

Prepare

make venv
source .venv/bin/activate

Run Server

python -m cogstream.cli.server

Run server and use Edge TPU if possible:

cogstream_tpu=True python -m cogstream.cli.server

Run Client

Classify all images in a directory ./images/ with a pre-trained mobilenet model:

python -m cogstream.cli.client --engine mobilenet  --source ./images/

Pass the --host <ip> flag if the server is running over the network.

Run emotion detection from camera

python -m cogstream.cli.client --engine ferplus --source camera

Setup individual ML capabilities

Using a camera as a client source

For camera streaming you need opencv on the client

pip install opencv-python

TensorFlow Lite

For the server hosting the ML models install tflite: https://www.tensorflow.org/lite/guide/python. For example on a Coral Dev Board (aarch64) with Python 3.5 run:

pip3 install https://dl.google.com/coral/python/tflite_runtime-2.1.0.post1-cp35-cp35m-linux_aarch64.whl

MXNet

For serving ferplus on will need mxnet.

pip install mxnet

For using the GPU you need CUDA and the mxnet cuda pip package. Make sure the CUDA version matches the CUDA version in the pip package of, e.g., mxnet-cu102:

cat /usr/local/cuda/version.txt
# OR
nvcc --version

# mxnet
pip install mxnet-cu<version>

cognitivexr/cogstream-py