Lewati ke konten
Kembali ke blog

asli bahasa inggris

Benchmark · Cross-platform

Selected public scenes tested on Mac, Windows and Linux

Explore image quality and cross-platform results from nine runs on three selected public scenes: Train, Dr Johnson and Playroom.

MakeWorlds team · Tested

Locomotives and tracks in the Train scene
Train · Mac · view-000057

Quality across three selected scenes

All three selected scenes improve on the cited paper baseline in PSNR, SSIM and LPIPS on each tested platform.

Ranges across three platforms · 30,000 steps · paper-comparison protocol
Δ = MakeWorlds − paper. Higher PSNR / SSIM and lower LPIPS are better.
Public scenePSNR ↑ / dBSSIM ↑LPIPS ↓
Train+0.108 – +0.185+0.0443 – +0.0459−0.0652 – −0.0643
Dr Johnson+0.404 – +0.508+0.0105 – +0.0134−0.0225 – −0.0169
Playroom+0.161 – +0.391+0.0077 – +0.0098−0.0158 – −0.0142

This article features three selected public scenes. Each range covers the Mac, Linux and Windows results for that scene using the paper-comparison protocol. Read the paper

Differences across platforms

Each selected scene ran on three platforms. The following table compares their test-set means using the separate cross-platform protocol.

Maximum minus minimum PSNR across platforms · cross-platform protocol
Public sceneCross-platform spread / dB
Train0.072
Dr Johnson0.104
Playroom0.230

All PSNR spreads in this round are within the registered 0.50 dB tolerance. The cross-platform and paper-comparison protocols are reported separately.

Test conditions and sources

Mac: Apple M4 Max; Linux and Windows: NVIDIA RTX 4090

All three scenes use Medium, source resolution and 30,000 training steps. The product solves its own cameras from the photographs before training and delivery.

Tanks and Temples · 3DGS

Explore a scene

From 301 photos to a 3D scene: the Train benchmark