holdenb/pycvss

A python module for the ULL High Performance Cloud Computing (HPCC) lab's Cloud Video Streaming Service (CVSS) architecture designed for research and service provision.

pycvss

This is a python module for the CVSS architecture designed for research and service provision 💻

Quick installation guide for the PYCVSS module ☕ 👓

pip install -r requirements.txt --user
pip install -e . --user

Pipenv can also be used to install the module and it's dependencies

pip install pipenv --user
pipenv install -dev
pipenv install -e .

## Open up a pipenv shell (this will enable your virtual environment)
pipenv shell
## Run scripts within the shell
## Exit the shell when finished
exit

Note: You will need to install both pytorch and ffmpeg:

• FFmpeg: https://github.com/adaptlearning/adapt_authoring/wiki/Installing-FFmpeg
• Pytorch: https://pytorch.org/get-started/locally/

Using a pre-trained SSD network for detection

Download a pre-trained network

• Currently, the following PyTorch models are available from Max deGroot & Ellis Brown:
  • SSD300 trained on VOC0712 (newest PyTorch weights)
    • https://s3.amazonaws.com/amdegroot-models/ssd300_mAP_77.43_v2.pth
  • SSD300 trained on VOC0712 (original Caffe weights)
    • https://s3.amazonaws.com/amdegroot-models/ssd_300_VOC0712.pth

Once downloaded, place the chosen training file within the ssd/training_files/ directory.

Running a service:

To list all available services:

py run_service.py services

Service commands take on the following format:

py run_service.py {service_name_subarg} {-arg}

Running the motioncapture service:

## To show the help menu for a specific service
py run_service.py motioncapture --help

## Running the service
py run_service.py motioncapture -i path/to/input_file.mp4 -o output_file.mp4

TODO

This is the current TODO list for improvements to the PYCVSS module:

• Still to come:
  • Initial implementation of Service Management object and module entry point
  • Initial implementation of motion capture and detection services
  • Implementation of FFMPEG calls (such as grayscale conversion, etc.) as services on the management object
  • Additional FFPEG bindings / process arg calls