DynGhost: Temporally-Modelled Transformer for Dynamic Ghost Imaging with Quantum Detectors
2026-05-11 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial Intelligence
AI summaryⓘ
The authors developed DynGhost, a new AI model that improves ghost imaging, which is a way to create pictures using only simple light detectors. Unlike earlier models, DynGhost can handle moving scenes by understanding how images change over time and deals better with the way real detectors count photons (tiny light particles). They trained their model using realistic simulations of these detectors to make it work well under real conditions. Tests showed that DynGhost does a better job than previous methods, especially when there is very little light or the scenes are dynamic.
Ghost imagingSingle-pixel detectorTransformer architectureTemporal coherencePoisson noiseAnscombe transformSuperconducting nanowire single-photon detectors (SNSPDs)Single-photon avalanche diodes (SPADs)Silicon photomultipliers (SiPMs)Photon-starved imaging
Authors
Vittorio Palladino, Ahmet Enis Cetin
Abstract
Ghost imaging reconstructs spatial information from a single-pixel bucket detector by correlating structured illumination patterns with scalar intensity measurements. While deep learning approaches have achieved promising results on static scenes, two critical limitations remain unaddressed: existing architectures fail to exploit temporal coherence across frames, leaving dynamic ghost imaging largely unsolved, and they assume additive Gaussian noise models that do not reflect the true Poissonian statistics of real single-photon hardware. We present DynGhost (Dynamic Ghost Imaging Transformer), a transformer architecture that addresses both limitations through alternating spatial and temporal attention blocks. Our quantum-aware training framework, based on physically accurate detector simulations (SNSPDs, SPADs, SiPMs) and Anscombe variance-stabilizing normalization, resolves the distribution shift that causes classical models to fail under realistic hardware constraints. Experiments across multiple benchmarks demonstrate that DynGhost outperforms both traditional reconstruction methods and existing deep learning architectures, with particular gains in dynamic and photon-starved settings.