1 of 9

Rich in information

Unlock insights with ML

Size

Volume

Unique processing requirements

Large amount of metadata

Complex and Challenging

Competitive edge

Images and Videos are Special.

https://tenor.com/view/deep-x-safety-compliance-ai-deep-x-warehousing-gif-gif-22102128

2 of 9

DEMO

Think back to the last time you created a video dataset for something

Train models to detect people and activity in videos captured in locations X, Y, Z

https://xnet.senstar.com/webhelp/Symphony/7.2/en/topic_concept/air_indoor_people_introduction.html

3 of 9

DIY Visual Data Management for Analytics

Databases

Storage

Ingest / ETL

Images / videos

+

Metadata

Image / video processing

Query scripts to build datasets

ML frameworks

Datasets

Graphical frontends for display

Embeddings

Models to deploy

Metadata

Images / videos

Labels

Datasets

Labels

Images / videos

Labeling frameworks

K-NN search libraries or in house scripts for similarity search

Recommendations

Cameras

Digital Assets/ Public Datasets

Labelers

4 of 9

Videos Add More Misery

Splitting into intervals can split interesting events across video snippets

Allocate a large enough VM to contain expected dataset

Create copies of interesting frames

Challenging to query, visualize, and debug

Days to prepare datasets at times, with tons of technical debt, and wasted resources

5 of 9

Can We Simplify Video (+ Image) Management and Access for Analytics?

Vishakha Gupta

vishakha@aperturedata.io

6 of 9

ApertureDB: Purpose-built Database for Visual Analytics

Query Engine (Orchestrator)

Implements the JSON-based native API

Query / update metadata or data

Capture snapshots of training data

On the fly visual data preprocessing

Find visual data using annotations

ML PIPELINES/ END USERS

Ingest Data

Train Model

Validate Model

Deploy Model

Built-in similarity matching for high-dimensional�feature vectors

SIMILARITY SEARCH

Native support for images, videos, and pre-processing operations

VISUAL DATA

Knowledge graph of user metadata, annotations, for advanced visual search

METADATA

ApertureDB

Label Data

7 of 9

It can be made simple – down to minutes

8 of 9

Simpler Data Pipeline Shifts Focus to ML / Data Science

Models to deploy

Recommendations

Images / videos +

Metadata

Annotations

ML Frameworks

Graphical frontends for display

Labeling Frameworks

Transform

Query datasets using keyword or similarity search

Cameras

Vendors

Labelers

9 of 9

Let’s chat during the poster session about your data challenges and how we are building ApertureDB!

team@aperturedata.io