MetaTele: Compact Refractive Metasurface Computational Telephoto Camera

† Co-first authors with equal contribution

Harshana Weligampola^1,†, Yuanrui Chen^1,†, Abhiram Gnanasambandam², Dilshan Godaliyadda²,
Hamid R. Sheikh², Stanley H. Chan¹, Qi Guo^1,*

¹Elmore Family School of Electrical and Computer Engineering, Purdue University ²Samsung Research America

Paper arXiv Code Dataset (mono) Dataset (color)

MetaTele overview. (a) The proposed MetaTele imaging system uses a hybrid refractive–metasurface assembly to form a compact telephoto architecture. (b) Two complementary raw measurements are captured sequentially: a structure image I_s under a narrow spectral band (sharp, monochrome) and a color cue I_c over the full visible spectrum (color but aberrated). A one-step diffusion model fuses these into a high-quality RGB telephoto image. (c–d) MetaTele achieves a telephoto ratio of 0.44 — the lowest reported for full-color RGB imaging.

Color Cue (I_c) vs. MetaTele Reconstruction

Move your mouse across the image to compare. Switch input type or scene below.

Color cue I_c (input) MetaTele output

Drag or hover to reveal. Right side always shows MetaTele's reconstruction.

Abstract

TL;DR — a metasurface + lens system that fits telephoto in a smartphone body, with a diffusion model doing the color reconstruction.

Smartphone cameras face fundamental form-factor constraints that limit their optical magnification, primarily due to the difficulty of reducing a lens assembly's telephoto ratio — the ratio between total track length (TTL) and effective focal length (EFL). Currently, conventional refractive optics struggle to achieve a telephoto ratio below 0.5 without requiring multiple bulky elements to correct optical aberrations. In this paper, we introduce MetaTele, a novel optics-algorithm co-design that breaks this bottleneck. MetaTele explicitly decouples the acquisition of scene structure and color information. First, it utilizes a compact refractive-metasurface optical assembly to capture a fine-detail structure image under a narrow wavelength band, inherently avoiding severe chromatic aberrations. Second, it captures a broadband color cue using the same optics; although this cue is heavily corrupted by chromatic aberrations, it retains sufficient spectral information to guide post-processing. We then employ a custom one-step diffusion model to computationally fuse these two raw measurements, successfully colorizing the structure image while correcting for system aberrations. We demonstrate a MetaTele prototype achieving an unprecedented telephoto ratio of 0.44 with a TTL of just 13 mm for RGB imaging, paving the way for DSLR-level telephoto capabilities within smartphone form factors.

0.44

Telephoto ratio (TTL/EFL)

13 mm

Total Track Length

30 mm

Effective Focal Length

2,650

Real-world dataset scenes

532 nm

Metasurface design wavelength

Optical System

A Galilean-telescope assembly: a refractive objective paired with a custom-designed metasurface eyepiece.

Optical model. MetaTele consists of a refractive objective lens (L) and a metasurface eyepiece (M) separated by distance m. The assembly magnifies the incident angle of incoming light waves, achieving telephoto compactness.

Metasurface Design

The metasurface is composed of Silicon Nitride (SiN) cylindrical nanopillars on a 300 nm × 300 nm grid, with a fixed height of 775 nm. The nanopillar radius uniquely determines the phase modulation at each nanocell, following a quadratic phase profile that acts as a diverging lens (focal length f = −2 mm). This design achieves near-diffraction-limited performance at 532 nm and provides higher tolerance to fabrication imperfections compared to purely refractive alternatives.

Two-Shot Capture

Structure image I_s: Captured with a 10 nm FWHM bandpass filter at 532 nm. Exploits the metasurface's design wavelength to achieve high spatial fidelity and near-diffraction-limited sharpness — but is monochrome.

Color cue I_c: Captured without the filter over the full visible spectrum. Contains rich chromatic information but suffers from severe chromatic aberrations due to the wavelength-dependent defocus of the metasurface.

Hardware prototype. (a) The assembled MetaTele system with Thorlabs AC050-008-A-ML objective lens (f = 7.5 mm, Ø5 mm), custom metasurface eyepiece, and Basler daA3840-45uc RGB sensor — all mounted on 5-axis precision stages. (b) Optical microscope image of the fabricated metasurface. Inset: SEM image at 13,000× magnification. (c) Experimentally measured PSFs for the structure image with MTF inset.

Phase delay profile at 532 nm

Transmittance profile at 532 nm

Simulation results. (a) Ray-tracing diagram illustrating the Galilean-telescope configuration at different field angles. (b) Modulation transfer functions (MTF) for the corresponding field angles. The system achieves near-diffraction-limited performance on axis.

Computational Model

A one-step diffusion model fuses structure and color to produce high-quality RGB telephoto images.

Computational framework. MetaTele uses a generator G_θ built on a one-step diffusion model Ω. The architecture adopts a variational encoder-decoder with a one-step diffusion module conditioned on text prompts (from I_s) and learned high-frequency feature embeddings (via adaptor network A). Training combines a data fidelity loss and the novel High-Frequency Variational Score Distillation (HF-VSD) loss.

HF-VSD Loss

The proposed High-Frequency Variational Score Distillation (HF-VSD) loss builds on standard VSD by applying a 2D high-pass filter in the frequency domain. This explicitly encourages the one-step diffusion model to recover fine texture details guided by the sharp structure image I_s, while preserving low-frequency color information from the color cue I_c.

Training Strategy

Encoder, decoder, and diffusion module are initialized from pre-trained Stable Diffusion weights. Only Low-Rank Adaptation (LoRA) parameters are fine-tuned, keeping computational cost low. The generator parameters θ and the HF-VSD parameters φ are updated alternately following the standard VSD framework.

Ablation: Effect of HF-VSD Loss

Comparing no VSD, standard VSD, our HF-VSD loss, and the ground truth on a representative crop.

w/o VSD

Standard VSD

HF-VSD (Ours)

Ground Truth

Results

Quantitative and qualitative comparisons against prior metasurface-based imaging systems.

System Comparison

MetaTele vs. prior metasurface-based cameras (Yang et al. 2022, Tseng et al. 2021, Pinilla et al. 2023) on simulated measurements from the Flickr2K dataset. Click a scene to compare.

Method	PSNR ↑	SSIM ↑	LPIPS ↓	DISTS ↓	FID ↓	NIQE ↓	MUSIQ ↑	MANIQA ↑	CLIPIQA ↑
Color cue	13.27	0.383	0.849	0.527	372.6	10.63	17.74	0.171	0.233
Yang et al.	15.01	0.443	0.543	0.339	192.4	5.99	37.02	0.203	0.240
Tseng et al.	24.56	0.803	0.301	0.194	125.0	4.75	46.98	0.238	0.393
Pinilla et al.	24.54	0.707	0.387	0.211	163.6	5.23	35.22	0.209	0.266
Ours	21.95	0.629	0.204	0.140	108.9	3.98	61.09	0.375	0.512

Best 2nd 3rd

PSF Comparison with Prior Work

Simulated PSFs of MetaTele and prior metasurface imaging systems at identical paraxial image heights. Inset numbers show Strehl ratios. MetaTele intentionally prioritizes sharpness at the 532 nm design wavelength, achieving the highest on-design Strehl ratio and near-uniform PSFs across the full field of view.

Optical Design Study

Simulation study. (a) Minimum telephoto ratios vs. number of optical elements under three scenarios: full-spectrum refractive, single-wavelength refractive, and hybrid refractive+metasurface. (b) Optimized assembly with 4 lenses + metasurface reaching telephoto ratio 0.17. (c) Optical performance vs. lateral perturbation: the hybrid system maintains high Strehl ratio even when the eyepiece is displaced by 0.02 mm, while the purely refractive counterpart degrades sharply.

Continuous Zoom Capability

MetaTele supports continuous zoom from 20 to 50 mm equivalent focal length by adjusting the separation between the objective and metasurface eyepiece, all without changing lenses.

Citation

If you find this work useful, please cite:

@misc{weligampola2025metatele,
  title         = {MetaTele: Compact Refractive Metasurface Computational Telephoto Camera},
  author        = {Weligampola, Harshana and Chen, Yuanrui and Gnanasambandam, Abhiram
                   and Godaliyadda, Dilshan and Sheikh, Hamid R. and Chan, Stanley H.
                   and Guo, Qi},
  year          = {2025},
  eprint        = {2604.07614},
  archivePrefix = {arXiv},
  primaryClass  = {eess.IV},
  url           = {https://arxiv.org/abs/2604.07614}
}

Funding: Samsung Research America Global Research Outreach & NSF Grant CCF-2431505 Fabrication: SNOChip Inc. Dataset: 2,650 real-world paired scenes (Flickr2K basis)