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 · UnderstandingEMIRATI RESEARCH. VISUAL INTELLIGENCE.
AI research for images, video and intelligent compression. Connecting the science of visual information with the companies building what comes next.
01 / EMIRATI AI RESEARCH
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 vision02 / WHAT WE EXPLORE
Three connected research directions, shaped around the needs of the application.
Explore how visual information can be processed, represented and understood while preserving the details that matter to people and AI systems.
Detail · Representation · UnderstandingStudy motion and the relationships between frames, with attention to temporal consistency, responsiveness and the demands of continuous visual streams.
Motion · Time · ContinuityInvestigate more efficient ways to represent images and video, balancing bitrate, quality, processing cost and the information needed by downstream applications.
Quality · Efficiency · Application03 / RESEARCH SPOTLIGHT
A selection of published research that frames the industry’s next questions, from learned image representations to practical neural video coding.
FOUNDATIONS / ICLR 2017
Ballé and colleagues connect learned image transforms with bitrate and reconstruction quality, establishing an important foundation for learned compression.
Read the paperVIDEO / CVPR 2019
DVC jointly optimizes learned motion and residual compression, extending the problem from individual images to predictive video coding.
Read the paperPRACTICAL SYSTEMS / CVPR 2025
DCVC-RT brings memory traffic and execution overhead into codec design, connecting compression research with the constraints of real-time operation.
Read the paperIndependent research, interpreted by Deep Refactor. Published results belong to their authors and remain subject to their test conditions.
04 / APPLICATION HORIZONS
Potential applications for collaborative research and evaluation, from connected infrastructure to the delivery of moving images.
DISTRIBUTED VISUAL SYSTEMS
Explore how distributed visual systems can retain useful information for traffic analysis, asset monitoring and event review within network and storage constraints.
Explore the possibilityQUALITY AT SCALE
Balance perceptual quality, bandwidth and processing demands across high-resolution content, live streams and diverse viewing devices.
Explore the possibilityDETAIL THAT MATTERS
Investigate processing choices that preserve task-relevant detail while accounting for the limits of cameras, edge devices and inspection workflows.
Explore the possibilityVARIABLE LINKS. USEFUL VISION.
Study adaptive visual pipelines for remote inspection and operator awareness over links with changing capacity and latency.
Explore the possibility05 / INDUSTRY OUTLOOK
The opportunity is where visual information meets a real constraint: a network, an archive, an edge device or an AI task.
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 sourceIn 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 sourceITU 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 source06 / FROM RESEARCH TO USE
Understand the use case, the information that matters and the operating constraints.
Develop research questions and test promising approaches through focused experiments.
Assess quality, efficiency, latency and application performance together.
Shape promising work around the systems and workflows in which it must operate.
LET’S THINK FURTHER
Tell us about your application, your technical priorities and the challenge you want to solve.
Start a technical conversationinfo@deeprefactor.ai