Kuree / scanner

Efficient video analysis at scale

Scanner: Efficient Video Analysis at Scale

Scanner is a system for efficient video processing and understanding at scale. Scanner provides a python API for expressing computations and a heterogeneous runtime for scheduling these computations onto clusters of machines with CPUs or GPUs.

• Install
• Running Scanner
• Tutorials & Examples
• Documentation
• Contributing

Scanner is an active research project, part of a collaboration between Carnegie Mellon and Stanford. Please contact Alex Poms and Will Crichton with questions.

Install

There are two ways to build and run Scanner on your machine:

• Docker
• From Source

Docker

First, install Docker. If you have a GPU and you're running on Linux, install nvidia-docker 1.0 and run:

pip install --upgrade nvidia-docker-compose
wget https://raw.githubusercontent.com/scanner-research/scanner/master/docker/docker-compose.yml
nvidia-docker-compose pull gpu
nvidia-docker-compose run --service-ports gpu /bin/bash

Otherwise, you should run:

pip install --upgrade docker-compose
wget https://raw.githubusercontent.com/scanner-research/scanner/master/docker/docker-compose.yml
docker-compose pull cpu
docker-compose run --service-ports cpu /bin/bash

If these commands were successful, you should now have bash session at the Scanner directory inside the docker container. To start processing some videos, check out Running Scanner

From Source

Follow the instructions at INSTALL to build Scanner from source. To start processing some videos, check out Running Scanner.

Running Scanner

Since Scanner programs are written using a high-level python API, running a Scanner program is as simple as executing a python script. Let's run a Scanner job now to find all the faces of people in a video (you can also use your own video if you have one on-hand). Run the following commands:

cd path/to/your/scanner/directory
# Download an example video (or use your own)
wget https://storage.googleapis.com/scanner-data/tutorial_assets/star_wars_heros.mp4
# Run the Scanner program
python examples/apps/face_detection/main.py star_wars_heros.mp4

You should see several progress bars indicating the video is being processed. When finished, there will be an mp4 file in your current directory called star_wars_heros_faces.mp4 with bounding boxes drawn over every face in the original video. Congratulations, you just ran your first Scanner program! Here's a few next steps:

• To learn how to start writing your own Scanner programs, dive into the API with the tutorials.
• To run other Scanner programs on your videos, check out the examples.
• If you're looking for a code reference, check out the documentation

Tutorials & Examples & How-To's

The tutorials and examples are located in the examples directory. Some of the examples include:

• Locate and recognize faces in a video
• Detect shots in a film
• Search videos by image

Documentation

TODO(apoms)

Contributing

TODO(apoms)