Deep Video Analytics    

by Akshay Bhat

Deep Video Analytics is a platform for indexing and extracting information from videos and images. With latest version of docker installed correctly, you can run Deep Video Analytics in minutes locally (even without a GPU) using a single command.

Architecture

Deep Video Analytics implements a client-server architecture pattern, where clients can access state of the server via a REST API. For uploading, processing data, training models, performing queries, i.e. mutating the state clients can send DVAPQL (Deep Video Analytics Processing and Query Language) formatted as JSON. The query represents a directed acyclic graph of operations.

Visual Data Network

A separate repository VisualDataNetwork/root maintains examples of DVAPQL scripts for performing tasks such as processing image dataset (e.g. COCO), Youtube videos, Twitch livestreams, training FAISS indexers etc.

Installation & Demo

Please visit https://www.deepvideoanalytics.com

Documentation

• For a quick overview we strongly recommend going through the presentation in readme.pdf

Experiments

• OCR example has been moved to /docs/experiments/ocr directory.

Code organization

• /client : Python client using DVA REST API
• /configs : ngnix config + default models + processing pipelines
• /deploy : Dockerfiles + Instructions for development, single machine deployment and scalable deployment with Kubernetes
• /docs : Documentation, tutorial and experiments
• /tests : Tests, Notebooks for interactive debugging andtest data
• /repos : Code copied from third party repos, e.g. Yahoo LOPQ, TF-CTPN etc.
• /server : dvalib + django server contains contains bulk of the code for UI, App and models.
• /logs : Empty dir for storing logs

Libraries present in this repository and their licenses

Library	Link to the license
YAD2K	MIT License
AdminLTE2	MIT License
FabricJS	MIT License
Facenet	MIT License
JSFeat	MIT License
MTCNN	MIT License
Insight Face	MIT License
CRNN.pytorch	MIT License
Original CRNN code by Baoguang Shi	MIT License
Object Detector App using TF Object detection API	MIT License
Plotly.js	MIT License
Text Detection CTPN	MIT License
SphereFace	MIT License
Segment annotator	BSD 3-clause
Youtube 8M feature extractor weights	Apache 2.0
LOPQ	Apache 2.0
Open Images Pre-trained network	Apache 2.0
Interval Tree	Apache 2.0

Libraries present in container (/root/thirdparty/)

Library	Link to the license
faiss	BSD + PATENTS License
dlib	Boost Software License

Additional libraries & frameworks

• FFmpeg (not linked, called via a Subprocess)
• Tensorflow
• OpenCV
• Numpy
• Pytorch
• Docker
• LMDB
• Nvidia-docker
• Docker-compose
• All packages in requirements.txt
• All dependancies installed in CPU Dockerfile & GPU Dockerfile

License & Copyright

Copyright 2016-2018, Akshay Bhat, All rights reserved.

Contact

Deep Video Analytics is nearing stable 1.0, we expect to release in Summer 2018. The license will be relaxed once a stable release version is reached. Please contact me for more information.