aiqbench analysis
This arXiv report summarizes the Mobile AI 2026 challenge, where participants built deep learning models that convert RAW Bayer sensor data into RGB images and run in under 100 ms on current Qualcomm and MediaTek (Arm Mali) mobile GPUs. The useful signal is the benchmark setup: it measures on-device runtime alongside image quality, using a large paired dataset from a Sony phone sensor and a 102MP medium-format reference camera. Readers building efficient vision models for phones should treat these results as a snapshot of what current mobile hardware can deliver, not as a general claim about ISP quality.
Learned Smartphone ISP on Mobile GPUs: Mobile AI 2026 Challenge Report
arXiv:2611.05008v1 Announce Type: cross Abstract: With the increasing importance of mobile camera quality in modern smartphones, growing attention has been directed toward computational photography and advanced image signal processing (ISP) algorithms aimed at enhancing various perceptual attributes of captured images. This creates a need for efficient deep learning models for this problem that are not only performant but can also be deployed on constrained mobile hardware. In this Mobile AI challenge, we address this problem and propose the participants to design fast solutions that can reconstruct RGB photos from RAW Bayer camera data while running on popular smartphone GPUs from Qualcomm and Arm. For this, the participants were provided with a large-scale Fujifilm UltraISP dataset consisting of RAW-RGB image pairs captured with the Sony IMX586 Quad Bayer mobile sensor and a professional 102MP medium format FujiFilm GFX100 camera. The runtime of all models was evaluated on the Snapdragon 8 Elite Gen 5 and Dimensity 9500 chipsets with the Adreno and Mali GPUs, respectively. Top solutions proposed in this challenge are able to reconstruct high-quality 5MP RGB image from raw data on both platforms under 100 milliseconds. A comprehensive description of the models developed in the challenge is provided in this paper.
Abstract:With the increasing importance of mobile camera quality in modern smartphones, growing attention has been directed toward computational photography and advanced image signal processing (ISP) algorithms aimed at enhancing various perceptual attributes of captured images. This creates a need for efficient deep learning models for this problem that are not only performant but can also be deployed on constrained mobile hardware. In this Mobile AI challenge, we address this problem and propose the participants to design fast solutions that can reconstruct RGB photos from RAW Bayer camera data while running on popular smartphone GPUs from Qualcomm and Arm. For this, the participants were provided with a large-scale Fujifilm UltraISP dataset consisting of RAW-RGB image pairs captured with the Sony IMX586 Quad Bayer mobile sensor and a professional 102MP medium format FujiFilm GFX100 camera. The runtime of all models was evaluated on the Snapdragon 8 Elite Gen 5 and Dimensity 9500 chipsets with the Adreno and Mali GPUs, respectively. Top solutions proposed in this challenge are able to reconstruct high-quality 5MP RGB image from raw data on both platforms under 100 milliseconds. A comprehensive description of the models developed in the challenge is provided in this paper.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2611.05008 [cs.CV]
(or arXiv:2611.05008v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2611.05008 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 296 words
Related coverage
Self-Supervised Graph Representation Learning for In-The-Wild Wearable and Smartphone based Emotion Recognition
arXiv · Aug 25, 2026
Bridging Search and CRM: Productionizing AI Product Research Agents for Customer Re-Engagement
arXiv · Aug 21, 2026
The best Garmin watches of 2026: Expert tested and reviewed
ZDNet · Aug 4, 2026
Nothing’s brand chief is fed up of ‘boring and generic’ smartphone ads—he shares his marketing strategy for taking on Samsung and Apple
Fortune · Oct 20, 2026