DOGNN is our independently developed technical standard for third parties building on our systems and devices. We are building a developer community around specifications, code and practice, exploring neural networks for more efficient embedded display pipelines.
Dynamic Optimize Graphic Neural Network
DOGNN provides the technical standard for third-party development on our systems and devices. The developer community will grow around understanding the specification, integration, hands-on code and technical exchange.
Technical scope, interface contracts and release plans depend on confirmed project materials. This site makes no claim of industry certification, open-standard status or open-source licensing.
Clarify the contracts developers follow.
Support development with code and tools.
Explain usage through concrete scenarios.
More efficient displays begin with less unnecessary work—and savings that exceed the cost of the neural network itself.
Visual change & compute allocation / Concept
Neural networks can contribute to image generation, reconstruction and update strategies: reducing base rendering work, reusing information across frames and allocating computation by need. Display transfers, hardware timing and synchronization must still meet device requirements. DOGNN explores net efficiency across the pipeline.
Generic pipeline: neural optimization may target image computation. This is not a published DOGNN architecture.
One approach reduces base rendering work, then reconstructs the image using a neural network and history. It helps only when reconstruction costs less than the rendering saved; text, fine lines and readings must remain accurate.
Consecutive frames share information. Learned decisions may help determine what to update or reuse, but must be compared with existing partial-refresh and caching methods to establish added value.
Processing budgets may follow visual change and interaction needs, with implementations suited to compute, memory and bandwidth constraints. Resources should serve useful work while visual quality and responsiveness remain acceptable.
Optimized total cost = remaining rendering + inference + data movement & synchronization
Neural computation is not free. The optimization holds only if it remains leaner and faster after additional costs are included.
Neural-rendering research demonstrates the potential of learned sampling, temporal information reuse and reconstruction. Its tasks and hardware differ from embedded display drivers. The references provide background, not DOGNN measurements or endorsements.
Neural Temporal Adaptive Sampling and Denoising ↗MNSS: Neural Supersampling for Real-Time Rendering on Mobile Devices ↗Expand a resource to see its status. No specification or SDK downloads are currently available.
The team has completed feasibility calculations indicating that approximately 50% lower video-driver-related resource use and 2–3× interaction response performance are achievable.
Compared with the team’s current LVGL implementation.
Achievable according to team calculations; applies to video-driver-related resource use.
Achievable according to team calculations; precise measurement definitions await formal documentation.
The approximately 50% resource saving and 2–3× interaction response gain are team feasibility-calculation conclusions, not published hardware measurements or a guarantee of simultaneous gains in all scenarios. Actual performance depends on hardware, workloads and test conditions.
Current LVGL implementation; version, configuration and compiler settings to be confirmed.
Overall performance, resources and response; definitions and collection methods to be confirmed.
Comparable hardware and workloads are proposed; models, resolution and repetitions to be confirmed.
The team has confirmed the feasibility-calculation conclusions. Detailed calculations and hardware test reports are not provided on this site and are distinguished by their publication status.
DOGNN stands for Dynamic Optimize Graphic Neural Network. It is our independently developed technical standard, intended to provide shared contracts for third parties building on our systems and devices.
The project is intended to operate under a foundation in France. This affiliation is in preparation; institutional information will follow confirmation.
LET’S BUILD WITH CLARITY.
DOGNN.org