LiDAR

LiDAR Scanner

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
🥇 Point Transformer 0.779 33.13 0.954 ✓ Certified Zhao et al., ICCV 2021
🥈 RandLA-Net 0.753 31.91 0.942 ✓ Certified Hu et al., CVPR 2020
🥉 PnP-ADMM 0.655 29.1 0.840 ✓ Certified ADMM + denoiser prior
4 Bilateral Filter 0.641 27.41 0.868 ✓ Certified Tomasi & Manduchi, ICCV 1998

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
🥇 Point Transformer + gradient 0.720
0.758
31.09 dB / 0.932
0.730
29.7 dB / 0.912
0.671
27.19 dB / 0.863
✓ Certified Zhao et al., ICCV 2021
🥈 Bilateral Filter + gradient 0.623
0.650
25.05 dB / 0.804
0.628
24.69 dB / 0.792
0.590
23.4 dB / 0.746
✓ Certified Tomasi & Manduchi, ICCV 1998
🥉 PnP-ADMM + gradient 0.618
0.679
26.23 dB / 0.838
0.605
23.41 dB / 0.747
0.569
21.9 dB / 0.686
✓ Certified Venkatakrishnan et al., 2013
4 RandLA-Net + gradient 0.610
0.740
30.07 dB / 0.918
0.593
23.18 dB / 0.738
0.496
19.95 dB / 0.596
✓ Certified Hu et al., CVPR 2020

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 Point Transformer + gradient 0.758 31.09 0.932
2 RandLA-Net + gradient 0.740 30.07 0.918
3 PnP-ADMM + gradient 0.679 26.23 0.838
4 Bilateral Filter + gradient 0.650 25.05 0.804
Spec Ranges (3 parameters)
Parameter Min Max Unit
timing_jitter -50.0 100.0 ps
beam_divergence -0.1 0.2 mrad
range_walk -1.0 2.0 cm
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 Point Transformer + gradient 0.730 29.7 0.912
2 Bilateral Filter + gradient 0.628 24.69 0.792
3 PnP-ADMM + gradient 0.605 23.41 0.747
4 RandLA-Net + gradient 0.593 23.18 0.738
Spec Ranges (3 parameters)
Parameter Min Max Unit
timing_jitter -60.0 90.0 ps
beam_divergence -0.12 0.18 mrad
range_walk -1.2 1.8 cm
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 Point Transformer + gradient 0.671 27.19 0.863
2 Bilateral Filter + gradient 0.590 23.4 0.746
3 PnP-ADMM + gradient 0.569 21.9 0.686
4 RandLA-Net + gradient 0.496 19.95 0.596
Spec Ranges (3 parameters)
Parameter Min Max Unit
timing_jitter -35.0 115.0 ps
beam_divergence -0.07 0.23 mrad
range_walk -0.7 2.3 cm
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

LiDAR (Light Detection and Ranging) measures distances by emitting laser pulses and timing the round-trip to the reflecting surface. Automotive LiDAR systems use rotating multi-beam scanners (e.g., Velodyne HDL-64E) or solid-state flash LiDAR to acquire 3D point clouds at 10-20 Hz. The forward model is simple time-of-flight: d = c*t/2. The resulting sparse point cloud requires densification, ground segmentation, and object detection. Primary challenges include sparse sampling, intensity variation with surface reflectivity, and rain/fog attenuation.

Principle

Light Detection and Ranging (LiDAR) measures distances by emitting laser pulses (905 nm or 1550 nm) and timing their return after reflection from the scene (time-of-flight: d = c·t/2). A scanning mechanism (rotating mirror, MEMS, or optical phased array) sweeps the beam to build a 3-D point cloud of the environment. Resolution depends on the beam divergence, scanning density, and pulse timing precision.

How to Build the System

Select a LiDAR sensor appropriate for the application: mechanical spinning (Velodyne VLP-16/128 for autonomous vehicles), solid-state (Livox, Ouster), or airborne (Leica ALS80 for terrain mapping). Mount rigidly and combine with an IMU and GNSS for georeferencing. Calibrate intrinsic parameters (beam angles, timing offsets, intensity response) and extrinsics (relative to vehicle coordinate frame). Process returns: first/last/full waveform for different applications.

Common Reconstruction Algorithms

  • Point cloud registration (ICP, NDT for multi-scan alignment)
  • Ground filtering and classification (progressive morphological filter)
  • SLAM (Simultaneous Localization and Mapping) with LiDAR
  • Object detection and segmentation (PointNet, PointPillars)
  • Surface reconstruction from point clouds (Poisson, ball-pivoting)

Common Mistakes

  • Multi-echo / multi-path reflections causing ghost points
  • Motion distortion in the point cloud from vehicle movement during one scan rotation
  • Incorrect calibration causing misalignment between LiDAR and camera data
  • Rain, fog, or dust causing false returns and reduced range
  • Near-range blind zone where the receiver is not sensitive to returns

How to Avoid Mistakes

  • Filter ghost points using intensity thresholds and multi-return analysis
  • Apply ego-motion compensation using IMU data to deskew each scan
  • Perform target-based or targetless calibration between LiDAR and other sensors
  • Use 1550 nm wavelength (eye-safe and less affected by rain) for outdoor applications
  • Account for minimum range specification; fuse with short-range sensors if needed

Forward-Model Mismatch Cases

  • The widefield fallback produces a 2D (64,64) image, but LiDAR produces a 1D or 3D point cloud of range measurements (r_i = c*t_i/2) — the output is a set of (x,y,z) points, not a blurred image
  • LiDAR measures distance by timing laser pulse round-trips, with angular scanning determining direction — the widefield spatial blur has no connection to time-of-flight distance measurement or angular scanning geometry

How to Correct the Mismatch

  • Use the LiDAR operator that models pulsed laser emission, scene reflection (surface albedo and geometry), and time-of-flight detection: range = c*delta_t/2 for each beam direction
  • Process the point cloud using registration (ICP), ground classification, or object detection algorithms that operate on the correct 3D range measurement format

Experimental Setup — Signal Chain

Experimental setup diagram for LiDAR Scanner

Experimental Setup

Instrument: Velodyne HDL-64E / Ouster OS1-128 / Livox Avia
Channels: 64
Range M: 120
Horizontal Fov Deg: 360
Vertical Fov Deg: 27
Horizontal Resolution Deg: 0.08
Rotation Rate Hz: 10
Wavelength Nm: 905
Points Per Second: 2200000
Dataset: KITTI, nuScenes, Waymo Open

Key References

  • Geiger et al., 'Are we ready for autonomous driving? The KITTI vision benchmark suite', CVPR 2012

Canonical Datasets

  • KITTI 3D object detection
  • nuScenes (1000 driving scenes)
  • Waymo Open Dataset

Spec DAG — Forward Model Pipeline

P(pulsed) → Σ(return) → D(g, η₁)

P Pulsed Laser (pulsed)
Σ Return Signal Integration (return)
D SPAD / APD (g, η₁)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
Δt timing_jitter Timing jitter (ps) 0 50
Δθ beam_divergence Beam divergence error (mrad) 0 0.1
ΔR range_walk Range walk error (cm) 0 1.0

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.