Lensless

Lensless (Diffuser Camera) Imaging

Join the Competition Contribute to Benchmark

Standard reconstruction benchmark — forward model perfectly known, no calibration needed. Score = 0.5 × clip((PSNR−15)/30, 0, 1) + 0.5 × SSIM

# Method Score PSNR (dB) SSIM Source
🥇 Uformer 0.768 33.5 0.920 ✓ Certified Wang et al., CVPR 2022
🥈 FlatNet 0.725 31.8 0.890 ✓ Certified Khan et al., IEEE TPAMI 2020
🥉 PnP-ADMM 0.603 27.5 0.790 ✓ Certified Monakhova et al., Opt. Express 2019
4 Wiener-ADMM 0.462 23.5 0.640 ✓ Certified Antipa et al., Optica 2018

Dataset: PWM Benchmark (4 algorithms)

Blind Reconstruction Challenge — forward model has unknown mismatch, must calibrate from data. Score = 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖)

# Method Overall Score Public
PSNR / SSIM
 Dev
PSNR / SSIM
 Hidden
PSNR / SSIM
 Trust Source
🥇 Uformer + gradient 0.701
0.787
32.28 dB / 0.946
0.694
27.65 dB / 0.873
0.622
23.92 dB / 0.766
✓ Certified Wang et al., CVPR 2022
🥈 FlatNet + gradient 0.645
0.739
30.02 dB / 0.917
0.612
23.55 dB / 0.752
0.584
22.9 dB / 0.727
✓ Certified Khan et al., IEEE TPAMI 2020
🥉 PnP-ADMM + gradient 0.611
0.652
25.13 dB / 0.806
0.623
24.29 dB / 0.779
0.558
22.37 dB / 0.706
✓ Certified Monakhova et al., Opt. Express 2019
4 Wiener-ADMM + gradient 0.523
0.538
20.51 dB / 0.623
0.526
20.77 dB / 0.635
0.506
20.07 dB / 0.602
✓ Certified Antipa et al., Optica 2018

Complete score requires all 3 tiers (Public + Dev + Hidden).

Join the competition →
Scoring: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖) PSNR 40% · SSIM 40% · Consistency 20%
Public 5 scenes

Full-access development tier with all data visible.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), spec ranges, ground truth (x_true), and true mismatch spec.

How to use: Load HDF5 → compare reconstruction vs x_true → check consistency → iterate.

What to submit: Reconstructed signals (x_hat) and corrected spec as HDF5.

Public Leaderboard
# Method Score PSNR SSIM
1 Uformer + gradient 0.787 32.28 0.946
2 FlatNet + gradient 0.739 30.02 0.917
3 PnP-ADMM + gradient 0.652 25.13 0.806
4 Wiener-ADMM + gradient 0.538 20.51 0.623
Spec Ranges (3 parameters)
Parameter Min Max Unit
diffuser_psf -5.0 10.0 %
sensor_distance -0.2 0.4 mm
wavelength -5.0 10.0 nm
View Details →
Dev 5 scenes

Blind evaluation tier — no ground truth available.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), and spec ranges only.

How to use: Apply your pipeline from the Public tier. Use consistency as self-check.

What to submit: Reconstructed signals and corrected spec. Scored server-side.

Dev Leaderboard
# Method Score PSNR SSIM
1 Uformer + gradient 0.694 27.65 0.873
2 PnP-ADMM + gradient 0.623 24.29 0.779
3 FlatNet + gradient 0.612 23.55 0.752
4 Wiener-ADMM + gradient 0.526 20.77 0.635
Spec Ranges (3 parameters)
Parameter Min Max Unit
diffuser_psf -6.0 9.0 %
sensor_distance -0.24 0.36 mm
wavelength -6.0 9.0 nm
Submit Reconstruction View Details →
Hidden 5 scenes

Fully blind server-side evaluation — no data download.

What you get & how to use

What you get: No data downloadable. Algorithm runs server-side on hidden measurements.

How to use: Package algorithm as Docker container / Python script. Submit via link.

What to submit: Containerized algorithm accepting y + H, outputting x_hat + corrected spec.

Hidden Leaderboard
# Method Score PSNR SSIM
1 Uformer + gradient 0.622 23.92 0.766
2 FlatNet + gradient 0.584 22.9 0.727
3 PnP-ADMM + gradient 0.558 22.37 0.706
4 Wiener-ADMM + gradient 0.506 20.07 0.602
Spec Ranges (3 parameters)
Parameter Min Max Unit
diffuser_psf -3.5 11.5 %
sensor_distance -0.14 0.46 mm
wavelength -3.5 11.5 nm
Submit Algorithm View Details →

Blind Reconstruction Challenge

Challenge

Given measurements with unknown mismatch and spec ranges (not exact params), reconstruct the original signal. A method must be evaluated on all three tiers for a complete score. Scored on a composite metric: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖).

Input

Measurements y, ideal forward model H, spec ranges

Output

Reconstructed signal x̂

About the Imaging Modality

Lensless imaging replaces the objective lens with a thin optical element (phase diffuser or coded mask) placed directly near the sensor. Scene light produces a multiplexed caustic pattern encoding the entire scene. The forward model is y = H * x + n where H is determined by the mask's phase profile and mask-to-sensor distance. Each scene point contributes across many sensor pixels, yielding a multiplexing advantage. Reconstruction solves a large-scale inverse problem via ADMM or FISTA with total-variation or learned priors.

Principle

Lensless (diffuser-cam) imaging replaces the imaging lens with a thin diffuser or coded mask placed directly before the sensor. The sensor records a multiplexed pattern (caustic or speckle) that encodes the 3-D scene. Computational reconstruction inverts the known point-spread function of the diffuser to recover the image, enabling an extremely compact, lightweight camera suitable for miniaturized or in-vivo applications.

How to Build the System

Place a thin diffuser (ground glass, engineered phase mask, or Scotch tape) at a fixed, small distance (~1-5 mm) from a bare sensor (CMOS, e.g., Sony IMX sensor). Precisely characterize the diffuser PSF by scanning a point source across the field of view. Mount rigidly to prevent any relative motion between diffuser and sensor. For 3-D reconstruction, the depth-dependent PSF must be calibrated at multiple axial planes.

Common Reconstruction Algorithms

  • ADMM (alternating direction method of multipliers) with TV regularization
  • Wiener deconvolution (fast, single-step but lower quality)
  • Gradient descent with learned priors (DiffuserCam, neural network prior)
  • Tikhonov-regularized least squares
  • Unrolled optimization networks (physics-informed deep learning)

Common Mistakes

  • Inaccurate PSF calibration causing reconstruction artifacts
  • Insufficient sensor dynamic range for the caustic intensity peaks
  • Motion between diffuser and sensor during capture invalidating the PSF model
  • Regularization too strong, over-smoothing fine details in the reconstruction
  • Ignoring the depth-dependence of the PSF when imaging 3-D scenes

How to Avoid Mistakes

  • Calibrate PSF carefully with a point source at the exact sample distance
  • Use HDR acquisition or high-bit-depth sensors to capture full caustic range
  • Rigidly bond the diffuser to the sensor; verify alignment stability
  • Tune regularization weight (e.g., via L-curve or cross-validation)
  • Calibrate PSF at multiple depths for 3-D scenes; use depth-varying reconstruction

Forward-Model Mismatch Cases

  • The widefield fallback uses a Gaussian PSF, but lensless cameras use a coded aperture (phase mask, diffuser, or amplitude mask) that creates a highly structured, non-Gaussian PSF — the caustic pattern is fundamentally different from a Gaussian
  • The lensless PSF encodes the scene through a known, shift-variant pattern — the widefield shift-invariant Gaussian blur does not capture the scene-dependent structure of the lensless measurement and produces incorrect reconstruction input

How to Correct the Mismatch

  • Use the lensless operator with the calibrated PSF of the specific coded aperture (measured from a point source or computed from the mask design): y = H * x, where H is the non-Gaussian, possibly shift-variant PSF
  • Reconstruct using Wiener deconvolution, ADMM with TV prior, or learned methods (FlatNet, PhlatCam) that use the correct coded-aperture PSF for the specific mask in use

Experimental Setup — Signal Chain

Experimental setup diagram for Lensless (Diffuser Camera) Imaging

Experimental Setup

Instrument: DiffuserCam / FlatCam prototype
Sensor: Raspberry Pi HQ Camera (Sony IMX477, 4056x3040)
Pixel Pitch Um: 1.55
Diffuser Type: optical diffuser (Luminit 0.5-deg) / coded mask
Diffuser To Sensor Mm: 2.5
Field Of View Deg: 40
Image Size: 2592x1944

Key References

  • Antipa et al., 'DiffuserCam: lensless single-exposure 3D imaging', Optica 5, 1-9 (2018)
  • Asif et al., 'FlatCam: Thin, Lensless Cameras Using Coded Aperture', IEEE TCI 3, 384-397 (2017)

Canonical Datasets

  • DiffuserCam lensless mirflickr dataset (Monakhova et al.)
  • PhlatCam benchmark (Boominathan et al., IEEE TPAMI 2022)

Spec DAG — Forward Model Pipeline

P(diffuser) → D(g, η₁)

P Diffuser Propagation (diffuser)
D Bare Sensor (g, η₁)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
ΔPSF diffuser_psf Diffuser PSF calibration error (%) 0 5.0
Δd sensor_distance Diffuser-sensor distance error (mm) 0 0.2
Δλ wavelength Wavelength mismatch (nm) 0 5.0

Credits System

40%
Platform Profit Pool
Revenue allocated to benchmark rewards
30%
Winner Share
Top algorithm receives from pool
$100
Min Withdrawal
Minimum payout threshold
Spec Primitives Reference (11 primitives)
P Propagation

Free-space or medium propagation kernel (Fresnel, Rayleigh-Sommerfeld).

M Mask / Modulation

Spatial or spatio-temporal amplitude modulation (coded aperture, SLM pattern).

Π Projection

Geometric projection operator (Radon transform, fan-beam, cone-beam).

F Fourier Sampling

Sampling in the Fourier / k-space domain (MRI, ptychography).

C Convolution

Shift-invariant convolution with a point-spread function (PSF).

Σ Summation / Integration

Summation along a physical dimension (spectral, temporal, angular).

D Detector

Sensor readout with gain g and noise model η (Gaussian, Poisson, mixed).

S Structured Illumination

Patterned illumination (block, Hadamard, random) applied to the scene.

W Wavelength Dispersion

Spectral dispersion element (prism, grating) with shift α and aperture a.

R Rotation / Motion

Sample or gantry rotation (CT, electron tomography).

Λ Wavelength Selection

Spectral filter or monochromator selecting a wavelength band.