EMIRATI RESEARCH. VISUAL INTELLIGENCE.

Intelligence.
In every frame.

AI research for images, video and intelligent compression. Connecting the science of visual information with the companies building what comes next.

DR / OPTICAL STUDIES01 — FRAME / FIELD
GENERATIVE VISUAL STUDY
BASED IN THE UNITED ARAB EMIRATESDiscover Deep Refactor

01 / EMIRATI AI RESEARCH

From the Emirates.
Towards wider horizons.

A deeper way to see. A clearer path to application.

Deep Refactor is an Emirati AI research and development company focused on image processing, video intelligence and compression. We work with companies building AI applications to connect visual research with practical engineering.

Our focus is the information inside the frame: how it is represented, what it must preserve, and how it can move through real systems more efficiently.

Discover our vision

02 / WHAT WE EXPLORE

Image. Motion. Information.

Three connected research directions, shaped around the needs of the application.

01 / RESEARCH DIRECTION

Image intelligence

Explore how visual information can be processed, represented and understood while preserving the details that matter to people and AI systems.

Detail · Representation · Understanding
02 / RESEARCH DIRECTION

Video in context

Study motion and the relationships between frames, with attention to temporal consistency, responsiveness and the demands of continuous visual streams.

Motion · Time · Continuity
03 / RESEARCH DIRECTION

Intelligent compression

Investigate more efficient ways to represent images and video, balancing bitrate, quality, processing cost and the information needed by downstream applications.

Quality · Efficiency · Application

03 / RESEARCH SPOTLIGHT

Ideas shaping what comes next.

A selection of published research that frames the industry’s next questions, from learned image representations to practical neural video coding.

Read our technical perspectives

FOUNDATIONS / ICLR 2017

Learning how to represent an image.

Ballé and colleagues connect learned image transforms with bitrate and reconstruction quality, establishing an important foundation for learned compression.

Read the paper

VIDEO / CVPR 2019

Bringing motion into the model.

DVC jointly optimizes learned motion and residual compression, extending the problem from individual images to predictive video coding.

Read the paper

PRACTICAL SYSTEMS / CVPR 2025

Designing a codec that can run well.

DCVC-RT brings memory traffic and execution overhead into codec design, connecting compression research with the constraints of real-time operation.

Read the paper

Independent research, interpreted by Deep Refactor. Published results belong to their authors and remain subject to their test conditions.

04 / APPLICATION HORIZONS

Visual efficiency, in the contexts that matter.

Potential applications for collaborative research and evaluation, from connected infrastructure to the delivery of moving images.

05 / INDUSTRY OUTLOOK

A new role for
every bit of video.

The opportunity is where visual information meets a real constraint: a network, an archive, an edge device or an AI task.

01 / ERICSSON

The uplink deserves more attention.

Ericsson’s June 2026 report found faster uplink than downlink growth at 43 of 55 measured service providers.

Our view: this strengthens the case for evaluating visual processing closer to capture, especially when applications continuously upload images or video.

Read the source
02 / MPEG

Machines change what quality means.

In January 2026, MPEG advanced Video Coding for Machines to Draft International Standard, with machine-task performance central to the compression objective.

Our view: AI application builders need to evaluate the information their models use, alongside what a human viewer sees.

Read the source
03 / ITU / JVET

The codec roadmap remains open.

ITU reports that JVET issued a joint call for proposals for compression beyond VVC in July 2026.

Our view: there is continuing space for research, careful benchmarking and integration work across established and emerging approaches.

Read the source
Explore applications and market outlook

06 / FROM RESEARCH TO USE

Start with the application. Build from evidence.

See our approach
01

Define

Understand the use case, the information that matters and the operating constraints.

02

Explore

Develop research questions and test promising approaches through focused experiments.

03

Evaluate

Assess quality, efficiency, latency and application performance together.

04

Translate

Shape promising work around the systems and workflows in which it must operate.

LET’S THINK FURTHER

What could your visual system do better?

Tell us about your application, your technical priorities and the challenge you want to solve.

Start a technical conversationinfo@deeprefactor.ai