FPP-ML-Bench is the first open-source, photorealistic synthetic dataset for single-shot fringe projection profilometry (FPP). It contains 15,600 fringe images generated using VIRTUS-FPP in NVIDIA Isaac Sim, created by aharoon and last updated in February 2026. The dataset is designed to enable standardized benchmarking of deep learning approaches for 3D depth reconstruction.
Use Cases
- Benchmarking deep learning models for 3D reconstruction based on synthetic fringe images.
- Systematic comparison of algorithmic approaches for fringe projection profilometry.
- Training and evaluating neural networks for single-shot depth estimation from structured light patterns.
Strengths
- Contains 15,600 total fringe images.
- Described as the first open-source, photorealistic synthetic dataset for this specific technique.
- Explicitly designed for standardized benchmarking and systematic comparison of methods.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- aharoon via Hugging Face.
- Collection Method
- Synthetically generated using VIRTUS-FPP in NVIDIA Isaac Sim.
- Time Range
- null
- Freshness
- Last updated 2026-02-03 05:37:31; freshness should be verified.
- Geography
- null