A new imaging method could give researchers a way to watch ultrafast events unfold from a single exposure, replacing the traditional need to repeat an experiment many times under carefully synchronized conditions. In a study published in Light: Science & Applications, Y. Lai and J. Liang report a technique called multi-spatiotemporal projection, which combines spatial and temporal optical encoding to capture rapidly changing scenes in compressed form. The approach is designed for phenomena that occur too quickly for ordinary cameras, including transient light propagation, laser–matter interactions, plasma dynamics, microscopic biological processes, and other events whose defining details can disappear in trillionths or billionths of a second. Instead of recording a complete image at every instant, the system deliberately mixes information across space and time, then reconstructs the hidden sequence computationally from one measurement.
The central challenge is that ultrafast imaging is not simply a matter of using a faster camera. A camera must collect enough photons, resolve fine spatial structure, and distinguish events separated by extremely short time intervals. Conventional high-speed methods often rely on repetitive sampling: the same event is measured again and again while the timing is shifted from one acquisition to the next. That strategy works for stable and reproducible processes, but it fails when the event is irreversible, random, evolving from shot to shot, or too valuable to repeat. Single-shot imaging avoids this limitation, but it creates a difficult information problem. A two-dimensional detector must somehow preserve the spatial appearance of a scene while also recording its temporal evolution. Lai and Liang’s framework addresses that problem by encoding different portions of the spatiotemporal signal into distinguishable measurement patterns.
In a multi-spatiotemporal projection system, the object is not viewed as a static picture. It is treated as a three-dimensional data cube containing two spatial dimensions and one time dimension. Optical components project this data cube through several complementary coding operations, causing information from different locations and moments to appear in a controlled, compressed arrangement on the detector. Spatial projection determines how positions are mixed, while temporal projection introduces a separate signature for changes occurring at different times. The recorded frame therefore contains overlapping contributions from many points in space and time. The overlap is intentional: the system trades direct visibility for encoded information, much as a compressed audio file preserves a complex sound using a compact representation rather than storing every detail in an unprocessed form.
The key technical advance is the simultaneous use of multiple projection dimensions rather than relying on a single type of temporal shearing or spatial modulation. In mathematical terms, the measurement can be described by a forward model that maps the unknown ultrafast scene to the detector image. Reconstruction then requires solving an inverse problem: given the compressed measurement and a calibrated description of the optical system, an algorithm estimates the sequence that most plausibly produced the observed pattern. Such inverse problems are often underdetermined, meaning that many possible scenes could generate similar measurements. The method therefore depends on carefully designed projections, accurate calibration, and computational reconstruction that exploits the structure of real scenes, such as correlations between neighboring pixels and gradual changes between adjacent time frames.
This architecture could be especially important for experiments in which the first shot permanently changes the sample or the conditions. A laser pulse striking a material, a chemical reaction initiated by an optical trigger, a shock wave crossing a microscopic target, or a biological process occurring in a living specimen may not be repeatable with identical timing and initial conditions. A conventional pump–probe experiment can reconstruct dynamics by averaging measurements across many cycles, but averaging can erase fluctuations and rare behavior. A single-shot system instead records the entire event in one pass. That makes it possible to study nonreproducible dynamics, capture transient asymmetries, and investigate the natural variation that would otherwise be hidden by repeated-measurement protocols.
The method also reflects a broader shift in scientific imaging: cameras are increasingly becoming measurement platforms paired with computational models. In ordinary photography, the goal is to preserve a scene as directly as possible. In compressed ultrafast imaging, the optics are designed to create a coded measurement that may look unintelligible without reconstruction. The detector does not need to resolve every temporal slice independently, which can reduce the hardware burden associated with ultrafast sensors. Instead, the optical system performs part of the information processing before the light reaches the camera, while algorithms complete the task afterward. This division between optical encoding and digital recovery can offer a flexible route toward high-speed imaging using detector technologies that are individually slower than the event being observed.
The quality of the recovered movie depends on more than computational power. Photon noise, detector noise, imperfect knowledge of the projection patterns, optical aberrations, and limited signal levels can all introduce errors. Fast phenomena may also contain abrupt changes that are difficult to reconstruct if an algorithm assumes excessive smoothness. For that reason, a practical implementation must balance temporal resolution, spatial resolution, field of view, sensitivity, and reconstruction stability. Increasing the number or diversity of projections can improve the separation of overlapping information, but it may also require more complicated optics or reduce the amount of light available in each encoded component. The significance of the reported approach lies in treating these trade-offs as a unified spatiotemporal design problem rather than optimizing spatial and temporal performance independently.
The prospect of recording an ultrafast movie from a single exposure has an immediate viral appeal because it turns invisible motion into something that can be visualized after the event. Yet the scientific value goes beyond striking images. A reconstructed sequence can reveal when and where energy moves through a system, how a wavefront changes shape, or how a microscopic structure responds during a fleeting transition. In fields such as optics and photonics, this may help researchers analyze light–matter interactions and the behavior of engineered materials. In plasma and high-energy-density physics, it could provide information about unstable structures that evolve too rapidly or unpredictably for repeated scans. In microscopy, the same principle could support observations where illumination must be minimized or where the specimen cannot survive multiple exposures.
Lai and Liang’s work positions multi-spatiotemporal projection as a bridge between optical engineering and computational imaging. Its underlying message is that ultrafast measurement does not always require a detector that independently resolves every instant. By designing how information is mixed before detection, a single camera exposure can carry a much richer record than its two-dimensional appearance suggests. The method will need to be tested across different light levels, scene complexities, and experimental environments before its full practical reach is established, but it offers a compelling strategy for capturing events that happen once and vanish forever. If refined and adopted broadly, single-shot compressed imaging could allow researchers to move from reconstructing idealized, repeatable dynamics toward observing the unpredictable real-time behavior of the fastest processes in nature.
Subject of Research: Single-shot ultrafast compressed imaging using multi-spatiotemporal optical projection and computational reconstruction.
Article Title: Multi-spatiotemporal projection enables single-shot ultrafast compressed imaging
Article References:
Lai, Y., Liang, J. Multi-spatiotemporal projection enables single-shot ultrafast compressed imaging.
Light Sci Appl 15, 346 (2026). https://doi.org/10.1038/s41377-026-02438-8
Image Credits: AI Generated
DOI: 10.1038/s41377-026-02438-8
Keywords: ultrafast imaging, single-shot imaging, compressed imaging, spatiotemporal projection, computational imaging, optical encoding, inverse reconstruction, high-speed cameras
Tags: advances in light science and applicationscompressed ultrafast imagingcomputational reconstruction in ultrafast imaginglaser–matter interaction visualizationmicroscopic biological process imagingmulti-spatiotemporal projectionnovel techniques for high-speed event captureplasma dynamics imagingsingle-shot ultrafast scene capturespatial and temporal optical encodingtransient light phenomena imagingultrafast imaging
