The Datakalab Difference

Industry leading compression, adaptation, speed and precision

Compression

90% Lossless compression so that you can update models OTA using minimal bandwidth.

Auto-adaption

Embedded self-learning without the need for annotation data. Always accurate, all of the time.

Acceleration

Datakalab strips out the redundant calculations in the algorithms so that they run as nimble and fast as possible. Precision without the baggage.

Private by Design

With everything running on the edge, regulatory risk or uncertainty is a thing of the past!

Slide right for base model detection ➡️

 ⬅️ Slide left for Datakalab optimised detection

Slide the slider above to compare the difference between a base model (image on left) and a Datakalab optimised model (image on right), both architectures are SSD MBNET v2 Lite using 320x320 images from the PIROPO database
(People in Indoor Rooms with Perspective and Omnidirectional cameras).
_________________
So until computers, chairs and humans are interchangeable, use Datakalab instead of open source models! Check out the full video comparison here.

The greatest precision to compute ratio in the industry

Minimal HW Requirements

CPU- 2.4GHz x86 processor (single core per feed)
PoE IP Cameras- Minimum 720p resolution and 7 FPS- MJPEG or RTSP Stream format
Internet Connectivity for StatisticsAny ethernet router capable of sending a ~50MB of data/month

9 research papers

3 patents pending

Algorithm-schema-structured-Data

6x faster inference

6x faster to implement

16x smaller footprint

3.75x less energy 

70% accuracy improvement

The Benefits

1

Smaller models, smaller machines and smaller costs:Thanks to our optimisations, deploy Datakalab on machines that are cheaper than GPUs and still readily available despite supply chain constraints.

2

Algorithms that dynamically adjust to changing lighting conditions automagically: It's not the same thing detecting objects at dawn than it is at noon or evening. Datakalab algorithms self adapt during the day to maintain optimal precision even when the lighting conditions change or are different between cameras.  

3

Compression that saves bandwidth: We remove the unnecessary weights in the models so that they can rapidly be updated and deployed without costing a fortune in bandwidth or compute.

4

Re-use existing cameras and hardware: You don't have to go out and get a new set of cameras that are stuck doing only one thing. Take advantage of your existing vision infrastructure and add counting, demographics or other use cases without having to add or cable additional hardware.

Have questions?

Tell us what's on your mind. We'd love to talk.