Datasets & Calibration

CVOS Lab provides public benchmark datasets for developing and auditing single-shot structured light 3D reconstruction networks (such as fringe-to-depth and fringe-to-phase mapping).

Benchmark Dataset

Dataset 1: Fringe Projection Dataset

A foundational dataset engineered for evaluating direct convolutional neural network models transforming standard sinusoidal fringe images directly to 3D depth shapes. Includes raw images and ground truth arrays.

Collection Date October 10, 2018
Total Samples 658 samples
Split Setup 620 Train, 38 Test
Total File Size 819.78 MB

Interactive mm Converter

Result: 0.00 mm

*Conversion rate: divide by 4.40235316045

Download dataset.zip (820MB)
Next-Gen SIDO Model

Dataset 2: SIDO Shape Reconstruction

A massive volumetric database built to train Single-Input Dual-Output (SIDO) frameworks. It provides training materials to convert a single fringe frame into two intermediate phase and modulation outputs before 3D mapping.

Collection Date April 12, 2021
Total Samples 1,523 samples
Access Type Private Sharing Link
Total File Size 6.67 GB

Interactive mm Converter

Result: 0.00 mm

*Conversion rate: divide by 2.86334144363

Download SIDO.zip (6.67GB)