NeRF

Neural Radiance Fields

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
🥇	NeRFactor2 NeRFactor2 Barron et al., NeurIPS 2024 35.85 dB SSIM 0.966 Checkpoint unavailable	0.831	35.85	0.966	✓ Certified	Barron et al., NeurIPS 2024
🥈	GaussianShader GaussianShader Wang et al., ICCV 2024 35.18 dB SSIM 0.960 Checkpoint unavailable	0.816	35.18	0.960	✓ Certified	Wang et al., ICCV 2024
🥉	3D-GS++ 3D-GS++ Kerbl et al., SIGGRAPH 2024 34.52 dB SSIM 0.952 Checkpoint unavailable	0.801	34.52	0.952	✓ Certified	Kerbl et al., SIGGRAPH 2024
4	3D-GS 3D-GS Kerbl et al., SIGGRAPH 2023 33.3 dB SSIM 0.940 Checkpoint unavailable	0.775	33.3	0.940	✓ Certified	Kerbl et al., SIGGRAPH 2023
5	2DGS 2DGS Huang et al., CVPR 2024 31.44 dB SSIM 0.936 Checkpoint unavailable	0.742	31.44	0.936	✓ Certified	Huang et al., CVPR 2024
6	NeRF NeRF Mildenhall et al., ECCV 2020 31.19 dB SSIM 0.933 Checkpoint unavailable	0.736	31.19	0.933	✓ Certified	Mildenhall et al., ECCV 2020
7	Instant-NGP Instant-NGP Muller et al., SIGGRAPH 2022 31.1 dB SSIM 0.905 Checkpoint unavailable	0.721	31.1	0.905	✓ Certified	Muller et al., SIGGRAPH 2022
8	Mesh-GS Mesh-GS Li et al., ECCV 2024 30.48 dB SSIM 0.924 Checkpoint unavailable	0.720	30.48	0.924	✓ Certified	Li et al., ECCV 2024
9	Mip-NeRF 360 Mip-NeRF 360 Barron et al., CVPR 2022 29.4 dB SSIM 0.844 Checkpoint unavailable	0.662	29.4	0.844	✓ Certified	Barron et al., CVPR 2022
10	Photogrammetry Photogrammetry Structure-from-Motion baseline 26.49 dB SSIM 0.845 Try in SpecLab →	0.614	26.49	0.845	✓ Certified	Structure-from-Motion baseline
11	COLMAP+MVS COLMAP+MVS Schonberger & Frahm, CVPR 2016 26.4 dB SSIM 0.730 Try in SpecLab →	0.555	26.4	0.730	✓ Certified	Schonberger & Frahm, CVPR 2016

Dataset: PWM Benchmark (11 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
🥇	NeRFactor2 + gradient NeRFactor2 + gradient Barron et al., NeurIPS 2024 Score 0.742 Correct & Reconstruct →	0.742	0.818 34.58 dB / 0.965	0.735 30.14 dB / 0.919	0.674 26.62 dB / 0.849	✓ Certified	Barron et al., NeurIPS 2024
🥈	3D-GS++ + gradient 3D-GS++ + gradient Kerbl et al., SIGGRAPH 2024 Score 0.712 Correct & Reconstruct →	0.712	0.777 32.27 dB / 0.946	0.688 27.15 dB / 0.862	0.671 27.09 dB / 0.860	✓ Certified	Kerbl et al., SIGGRAPH 2024
🥉	GaussianShader + gradient GaussianShader + gradient Wang et al., ICCV 2024 Score 0.699 Correct & Reconstruct →	0.699	0.783 32.45 dB / 0.947	0.711 28.02 dB / 0.881	0.603 23.34 dB / 0.744	✓ Certified	Wang et al., ICCV 2024
4	Instant-NGP + gradient Instant-NGP + gradient Muller et al., SIGGRAPH 2022 Score 0.657 Correct & Reconstruct →	0.657	0.748 29.58 dB / 0.910	0.623 23.91 dB / 0.765	0.600 23.68 dB / 0.757	✓ Certified	Muller et al., SIGGRAPH 2022
5	3D-GS + gradient 3D-GS + gradient Kerbl et al., SIGGRAPH 2023 Score 0.654 Correct & Reconstruct →	0.654	0.761 31.19 dB / 0.933	0.638 25.05 dB / 0.804	0.563 22.19 dB / 0.698	✓ Certified	Kerbl et al., SIGGRAPH 2023
6	2DGS + gradient 2DGS + gradient Huang et al., CVPR 2024 Score 0.639 Correct & Reconstruct →	0.639	0.733 29.49 dB / 0.909	0.598 23.12 dB / 0.736	0.585 23.18 dB / 0.738	✓ Certified	Huang et al., CVPR 2024
7	NeRF + gradient NeRF + gradient Mildenhall et al., ECCV 2020 Score 0.636 Correct & Reconstruct →	0.636	0.726 28.99 dB / 0.900	0.602 23.31 dB / 0.743	0.580 22.37 dB / 0.706	✓ Certified	Mildenhall et al., ECCV 2020
8	COLMAP+MVS + gradient COLMAP+MVS + gradient Schonberger & Frahm, CVPR 2016 Score 0.626 Correct & Reconstruct →	0.626	0.653 24.64 dB / 0.791	0.611 24.22 dB / 0.776	0.613 24.21 dB / 0.776	✓ Certified	Schonberger & Frahm, CVPR 2016
9	Photogrammetry + gradient Photogrammetry + gradient Structure-from-Motion baseline Score 0.593 Correct & Reconstruct →	0.593	0.631 24.41 dB / 0.783	0.585 22.65 dB / 0.717	0.562 22.61 dB / 0.715	✓ Certified	Structure-from-Motion baseline
10	Mesh-GS + gradient Mesh-GS + gradient Li et al., ECCV 2024 Score 0.548 Correct & Reconstruct →	0.548	0.713 28.26 dB / 0.886	0.512 20.36 dB / 0.616	0.419 17.55 dB / 0.478	✓ Certified	Li et al., ECCV 2024
11	Mip-NeRF 360 + gradient Mip-NeRF 360 + gradient Barron et al., CVPR 2022 Score 0.541 Correct & Reconstruct →	0.541	0.717 27.69 dB / 0.874	0.497 19.45 dB / 0.572	0.409 16.63 dB / 0.432	✓ Certified	Barron et al., CVPR 2022

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	NeRFactor2 + gradient	0.818	34.58	0.965
2	GaussianShader + gradient	0.783	32.45	0.947
3	3D-GS++ + gradient	0.777	32.27	0.946
4	3D-GS + gradient	0.761	31.19	0.933
5	Instant-NGP + gradient	0.748	29.58	0.91
6	2DGS + gradient	0.733	29.49	0.909
7	NeRF + gradient	0.726	28.99	0.9
8	Mip-NeRF 360 + gradient	0.717	27.69	0.874
9	Mesh-GS + gradient	0.713	28.26	0.886
10	COLMAP+MVS + gradient	0.653	24.64	0.791
11	Photogrammetry + gradient	0.631	24.41	0.783

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
camera_pose	-1.0	2.0	mm/deg
focal_length	-5.0	10.0	pixels
distortion	-0.01	0.02
exposure	-10.0	20.0	%

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	NeRFactor2 + gradient	0.735	30.14	0.919
2	GaussianShader + gradient	0.711	28.02	0.881
3	3D-GS++ + gradient	0.688	27.15	0.862
4	3D-GS + gradient	0.638	25.05	0.804
5	Instant-NGP + gradient	0.623	23.91	0.765
6	COLMAP+MVS + gradient	0.611	24.22	0.776
7	NeRF + gradient	0.602	23.31	0.743
8	2DGS + gradient	0.598	23.12	0.736
9	Photogrammetry + gradient	0.585	22.65	0.717
10	Mesh-GS + gradient	0.512	20.36	0.616
11	Mip-NeRF 360 + gradient	0.497	19.45	0.572

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
camera_pose	-1.2	1.8	mm/deg
focal_length	-6.0	9.0	pixels
distortion	-0.012	0.018
exposure	-12.0	18.0	%

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	NeRFactor2 + gradient	0.674	26.62	0.849
2	3D-GS++ + gradient	0.671	27.09	0.86
3	COLMAP+MVS + gradient	0.613	24.21	0.776
4	GaussianShader + gradient	0.603	23.34	0.744
5	Instant-NGP + gradient	0.600	23.68	0.757
6	2DGS + gradient	0.585	23.18	0.738
7	NeRF + gradient	0.580	22.37	0.706
8	3D-GS + gradient	0.563	22.19	0.698
9	Photogrammetry + gradient	0.562	22.61	0.715
10	Mesh-GS + gradient	0.419	17.55	0.478
11	Mip-NeRF 360 + gradient	0.409	16.63	0.432

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
camera_pose	-0.7	2.3	mm/deg
focal_length	-3.5	11.5	pixels
distortion	-0.007	0.023
exposure	-7.0	23.0	%

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

Neural radiance fields (NeRF) represent a 3D scene as a continuous volumetric function F(x,y,z,theta,phi) -> (RGB, sigma) parameterized by a multi-layer perceptron that maps 5D coordinates (position + viewing direction) to color and volume density. Novel views are synthesized by marching camera rays through the volume and integrating color weighted by transmittance using quadrature. Training optimizes the MLP weights to minimize photometric loss between rendered and observed images. Primary challenges include slow training/rendering, view-dependent effects, and the need for accurate camera poses (from COLMAP).

Principle

Neural Radiance Fields (NeRF) represent a 3-D scene as a continuous volumetric function F(x,y,z,θ,φ) → (RGB, σ) parameterized by a multi-layer perceptron (MLP). The network maps 3-D position and viewing direction to color and volume density. Novel views are synthesized by differentiable volume rendering along camera rays, and the network is trained by minimizing photometric loss against a set of posed 2-D images.

How to Build the System

Capture 50-200 images of a scene from diverse viewpoints using a calibrated camera (known intrinsics) or estimate camera poses with COLMAP structure-from-motion. Images should cover the scene uniformly. Train a NeRF MLP (typically 8 layers, 256 units, with positional encoding of input coordinates) on a GPU (≥12 GB VRAM). Training takes 12-48 hours on a single V100. Use mip-NeRF, Instant-NGP, or TensoRF for faster convergence.

Common Reconstruction Algorithms

Vanilla NeRF (MLP + positional encoding)
Instant-NGP (multi-resolution hash encoding, minutes training)
mip-NeRF (anti-aliased cone tracing)
Nerfacto (nerfstudio default combining multiple improvements)
TensoRF (tensor factorization for compact radiance fields)

Common Mistakes

Insufficient camera pose accuracy (SfM failure) causing blurry results
Too few input views or views clustered in a narrow angular range
Training only at one scale without mip-NeRF, causing aliasing at novel distances
Floater artifacts in empty space from insufficient regularization
Very slow training and rendering with vanilla NeRF (hours to train, seconds per frame)

How to Avoid Mistakes

Verify COLMAP pose estimation quality; add more images if registration fails
Capture views uniformly around the scene; include close-up and distant views
Use mip-NeRF or multi-scale training for scale consistency
Add distortion loss or density regularization to eliminate floater artifacts
Use Instant-NGP or 3D Gaussian Splatting for real-time rendering requirements

Forward-Model Mismatch Cases

The widefield fallback processes a single 2D (64,64) image, but NeRF renders multiple views of a 3D scene from a volumetric radiance field — output shape (n_views, H, W) represents images from different camera poses
NeRF is fundamentally nonlinear (volume rendering integral: C(r) = integral of T(t)*sigma(t)*c(t) dt along each ray) — the widefield linear blur cannot model view-dependent appearance, occlusion, or 3D geometry

How to Correct the Mismatch

Use the NeRF operator that performs differentiable volume rendering: for each pixel, cast a ray through the volumetric density/color field and integrate transmittance-weighted radiance
Optimize the 3D radiance field (MLP or voxel grid) to minimize photometric loss across all training views using the correct volume rendering equation as the forward model

Experimental Setup — Signal Chain

Experimental setup diagram for Neural Radiance Fields (NeRF)

Reconstruction Gallery — 4 Scenes × 3 Scenarios

Method: CPU_baseline | Mismatch: nominal (nominal=True, perturbed=False)

I Ideal II Mismatch (uncorrected) III Oracle correction

I — Ideal

Ground Truth

Measurement

Reconstruction

21.32958707867516 dB SSIM 0.8106553265548637

II — Mismatch

Ground Truth

Measurement

Reconstruction

33.60904208582102 dB SSIM 0.9287468345470429

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

46.07103075919076 dB SSIM 0.9895611557693481

I — Ideal

Ground Truth

Measurement

Reconstruction

21.535410122278478 dB SSIM 0.8153944981857855

II — Mismatch

Ground Truth

Measurement

Reconstruction

33.795250092263906 dB SSIM 0.92988497286129

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

46.267163301535334 dB SSIM 0.9901547315559387

I — Ideal

Ground Truth

Measurement

Reconstruction

20.212485446945596 dB SSIM 0.7912008845696258

II — Mismatch

Ground Truth

Measurement

Reconstruction

32.89617634297629 dB SSIM 0.925130273882866

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

46.424797644481366 dB SSIM 0.9898016602239609

I — Ideal

Ground Truth

Measurement

Reconstruction

22.403015763739038 dB SSIM 0.8189964130944609

II — Mismatch

Ground Truth

Measurement

Reconstruction

34.57071232223057 dB SSIM 0.9281706045341491

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

46.88024032085963 dB SSIM 0.989828976278305

Mean PSNR Across All Scenes

21.370124602909566 dB

I — Ideal

33.71779521082294 dB

II — Mismatch

46.410808006516774 dB

III — Oracle

Degradation: 12.3 dB Recovery: +12.7 dB

Per-scene PSNR breakdown (4 scenes)

Scene	I (PSNR)	I (SSIM)	II (PSNR)	II (SSIM)	III (PSNR)	III (SSIM)
scene_00	21.32958707867516	0.8106553265548637	33.60904208582102	0.9287468345470429	46.07103075919076	0.9895611557693481
scene_01	21.535410122278478	0.8153944981857855	33.795250092263906	0.92988497286129	46.267163301535334	0.9901547315559387
scene_02	20.212485446945596	0.7912008845696258	32.89617634297629	0.925130273882866	46.424797644481366	0.9898016602239609
scene_03	22.403015763739038	0.8189964130944609	34.57071232223057	0.9281706045341491	46.88024032085963	0.989828976278305
Mean	21.370124602909566	0.8090617806011839	33.71779521082294	0.927983171456337	46.410808006516774	0.9898366309568881

Experimental Setup

Training Views: 100

Test Views: 200

Image Resolution: 800x800

Scene Type: object-centric 360 deg (Blender synthetic)

Architecture: positional encoding + MLP (8 layers, 256 hidden)

Training Iterations: 200000

Batch Size Rays: 4096

Evaluation: PSNR / SSIM / LPIPS

Dataset: Blender Synthetic (8 scenes), LLFF (8 forward-facing)

Key References

Mildenhall et al., 'NeRF: Representing scenes as neural radiance fields for view synthesis', ECCV 2020
Muller et al., 'Instant Neural Graphics Primitives (Instant-NGP)', SIGGRAPH 2022

Canonical Datasets

NeRF Blender Synthetic (8 scenes)
LLFF (8 forward-facing scenes)
Mip-NeRF 360 (9 unbounded scenes)

Spec DAG — Forward Model Pipeline

Π(ray) → Σ(volume) → D(g, η₁)

Π Ray Casting (ray)

Σ Volume Rendering Integral (volume)

D Camera (g, η₁)

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
ΔT	camera_pose	Camera pose error (mm / deg)	1.0
Δf	focal_length	Focal length error (pixels)	5.0
Δk	distortion	Radial distortion coefficient error	0.01
ΔE	exposure	Exposure variation (%)	10.0

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.