Answers are grounded in this paper and cite their sources.
Real-world scenes can span more than 100 dB of luminance, while conventional image sensors typically capture less than 90 dB in a single exposure. This gap causes bright regions to saturate and dark regions to lose detail. Existing cameras extend the dynamic range by combining multiple sequentially captured exposures, but this increases total acquisition time and produces artifacts from scene or camera motion. In this work, we translate multi-exposure acquisitions from the temporal domain to the optical domain. A 2×3 metalens array with engineered transmittance across the full visible spectrum simultaneously captures six differently attenuated exposures in a single snapshot, which extends the dynamic range by 23 dB over a conventional single-aperture camera. However, each metalens in the array produces different, spatially varying blur and aberrations across the image, making the six measurements difficult to restore, align, and fuse into a high dynamic range (HDR) image. Correcting these degradations and performing HDR fusion typically requires GPU-intensive processing, which undermines the size and efficiency advantages of compact meta-optics. We overcome this computational bottleneck using content- and position-adaptive filters and a lightweight image restoration network. The complete optical-computational pipeline runs directly on an NVIDIA Jetson Orin Nano with 8 GB of memory, with network inference running at 2.7 fps at 256×256 resolution and 8.5 fps at 128×128 resolution, bringing HDR reconstruction within the compute and memory constraints of mobile-class hardware. Experiments on real-world high-contrast, low-light, and fast-moving scenes demonstrate robust reconstruction under conditions that challenge sequential multi-exposure capture. Together, these results move computational metalens imaging from GPU-intensive offline reconstruction toward compact, edge-deployable cameras.
No takes yet. Share an insight, caveat, or question.
Chakravarthula et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: