Update README.md

README.md

@@ -5,4 +5,183 @@ task_categories:
 - feature-extraction
 language:
 - en
----
+---
+
+# FPP-ML-Bench: Fringe Projection Profilometry Benchmarking Dataset
+
+[![arXiv](https://img.shields.io/badge/arXiv-2601.08900-b31b1b.svg)](https://arxiv.org/abs/2601.08900)
+[![SPIE Photonics West](https://img.shields.io/badge/SPIE%20Photonics%20West-Presentation-blue)](https://spie.org/photonics-west/presentation/Comprehensive-machine-learning-benchmarking-for-fringe-projection-profilometry-with-photorealistic/13904-1)
+[![GitHub](https://img.shields.io/badge/GitHub-Code-green)](https://github.com/AnushLak/fpp-ml-bench)
+
+The first open-source, photorealistic synthetic dataset for single-shot fringe projection profilometry (FPP), generated using [VIRTUS-FPP](https://arxiv.org/abs/2509.22685) in NVIDIA Isaac Sim. This dataset was created to enable standardized benchmarking and systematic comparison of deep learning approaches for single-shot 3D depth reconstruction from fringe patterns.
+
+## Dataset Summary
+
+| Property | Value |
+|----------|-------|
+| Total fringe images | 15,600 |
+| Depth reconstructions | 300 |
+| Objects | 50 |
+| Viewpoints per object | 6 |
+| Resolution | 960 × 960 pixels |
+| Measurement range | 1.5–1.8 m |
+| Ground truth method | 18-step phase shifting + Gray-code unwrapping |
+| Train / Val / Test split | 240 / 30 / 30 (object-level) |
+
+## Loading the Dataset
+
+```python
+import scipy.io as sio
+from PIL import Image
+import numpy as np
+
+# Load a fringe image
+fringe = np.array(Image.open("train/fringe/sample.png").convert("L"), dtype=np.float32) / 255.0
+
+# Load the corresponding ground truth depth map
+mat = sio.loadmat("train/depth/sample.mat")
+
+# Available depth keys depending on normalization:
+depth_raw                = mat["depthMap"]              # Raw depth in mm
+depth_global_normalized  = mat["depthMapMeters"]       # Depth in meters (raw / 1000)
+depth_individual_normalized = mat["depthMapNormalized"] # Per-sample [0, 1] normalized
+```
+
+### Denormalizing Individual Normalized Depth
+
+Each sample has stored normalization parameters to recover metric depth:
+
+```python
+# Load normalization parameters
+params = sio.loadmat("info_depth_params/train/depth/sample.mat")
+depth_min = float(params["depth_min"])
+depth_max = float(params["depth_max"])
+
+# Recover depth in mm from [0, 1] normalized prediction
+depth_mm = depth_normalized * (depth_max - depth_min) + depth_min
+```
+
+## Dataset Structure
+
+```
+fpp-ml-bench/
+├── train/                          # 240 samples (40 objects × 6 viewpoints)
+│   ├── fringe/                     # Input fringe pattern images (.png)
+│   └── depth/                      # Ground truth depth maps (.mat)
+├── val/                            # 30 samples (5 objects × 6 viewpoints)
+│   ├── fringe/
+│   └── depth/
+├── test/                           # 30 samples (5 objects × 6 viewpoints)
+│   ├── fringe/
+│   └── depth/
+└── info_depth_params/              # Per-sample normalization parameters
+    ├── train/depth/                # depth_min, depth_max per sample (.mat)
+    ├── val/depth/
+    └── test/depth/
+```
+
+### File Formats
+
+- **Fringe images** (`.png`): Single-channel 8-bit grayscale, 960×960 pixels. Each image is the first frame of an 18-step phase-shifting sequence.
+- **Depth maps** (`.mat`): MATLAB files containing three depth representations per sample:
+  - `depthMap`: Raw depth in millimeters
+  - `depthMapMeters`: Global normalized (raw / 1000, in meters)
+  - `depthMapNormalized`: Individual normalized to [0, 1] using per-sample min/max
+- **Normalization parameters** (`.mat`): Contains `depth_min` and `depth_max` for each sample, required to recover metric depth from individual normalized predictions.
+
+### Depth Normalization Strategies
+
+The dataset includes three depth representations to support systematic evaluation of data representation strategies. Our benchmarking study found these lead to dramatically different model performance:
+
+| Strategy | Key (`mat` file) | Range | Performance (Object MAE) |
+|----------|-----------------|-------|--------------------------|
+| Raw | `depthMap` | 0–2000 mm | 148.07 mm |
+| Global normalized | `depthMapMeters` | 0–2 m | 82.49 mm |
+| Individual normalized | `depthMapNormalized` | [0, 1] | 16.20 mm |
+
+Individual normalization decouples shape from scale, achieving 9.1× improvement over raw depth. See the [paper](https://arxiv.org/abs/2601.08900) for full analysis.
+
+## Data Acquisition
+
+### Virtual FPP System
+
+All data were generated using [VIRTUS-FPP](https://arxiv.org/abs/2509.22685), a physics-based virtual FPP system built in NVIDIA Isaac Sim. The pipeline integrates OptiX ray tracing for photorealistic rendering, PhysX for physics, and USD for 3D scene composition. The projector is modeled via the inverse camera model, enabling accurate fringe projection at arbitrary distances without hardware constraints.
+
+| Parameter | Value |
+|-----------|-------|
+| Camera focal length | 50 cm |
+| Camera resolution | 960 × 960 pixels |
+| Projector intensity | 40 nits |
+| Projector offset | 0.1 m below, 0.125 m left of camera |
+| Stereo reprojection error | 0.056 pixels |
+| Projector reprojection error | 0.049 pixels |
+
+### Objects
+
+50 objects were sourced from two datasets:
+- **YCB Object Dataset**: Household objects (containers, tools, food items)
+- **NVIDIA Physical AI Warehouse**: Industrial components
+
+Objects span cylindrical containers, rectangular boxes, complex shapes (power drills, sprayguns), and industrial components, providing diversity in surface characteristics and morphological complexity.
+
+All objects use consistent matte material properties: roughness = 0.95, specular = 0.15, AO-to-diffuse = 0.95.
+
+### Multi-View Acquisition
+
+Each object was rotated about the vertical axis in 60° increments, yielding 6 viewpoints per object with approximately 50% overlap between adjacent views:
+
+$$R_z(\theta_i) = \begin{bmatrix} \cos\theta_i & -\sin\theta_i & 0 \\ \sin\theta_i & \cos\theta_i & 0 \\ 0 & 0 & 1 \end{bmatrix}, \quad \theta_i = i \cdot 60°, \quad i = 0,1,...,5$$
+
+### Ground Truth Generation
+
+Ground truth depth maps were generated using an 18-step phase-shifting sequence combined with Gray-code temporal unwrapping and triangulation. This provides perfect ground truth geometry free from the measurement errors inherent to physical systems.
+
+## Train/Val/Test Split
+
+The dataset is split at the **object level** to ensure evaluation on completely unseen geometries:
+
+| Split | Objects | Samples | Percentage |
+|-------|---------|---------|------------|
+| Train | 40 | 240 | 80% |
+| Val | 5 | 30 | 10% |
+| Test | 5 | 30 | 10% |
+
+**Important:** No object appears in more than one split. This prevents the model from memorizing object shapes seen during training and forces generalization to novel geometries.
+
+## Intended Uses
+
+- Benchmarking deep learning architectures for single-shot FPP depth estimation
+- Evaluating data representation and loss function strategies for fringe-to-depth learning
+- Developing and validating learning-based 3D reconstruction methods
+- Research into the fundamental limitations of single-shot depth recovery from structured light
+
+## Limitations
+
+- **Synthetic only**: All data is rendered in simulation. Domain gap to real-world FPP systems has not been characterized. See [paper](https://arxiv.org/abs/2601.08900) for discussion and future work on sim-to-real transfer.
+- **Material properties**: All objects use identical matte materials. Specular, translucent, or highly reflective surfaces are not represented.
+- **Single projector pattern**: Only the first fringe image from each 18-step sequence is used as model input. The remaining 17 patterns are used solely for ground truth generation.
+- **Fixed measurement range**: All objects are scanned at 1.5–1.8 m. Performance at other distances is unknown.
+
+## Citation
+
+If you use this dataset in your research, please cite:
+
+```bibtex
+@article{lakshman2026comprehensive,
+  title={Comprehensive Machine Learning Benchmarking for Fringe Projection Profilometry with Photorealistic Synthetic Data},
+  author={Lakshman S, Anush and Haroon, Adam and Li, Beiwen},
+  journal={arXiv preprint arXiv:2601.08900},
+  year={2026}
+}
+
+@article{haroon2025virtus,
+  title={VIRTUS-FPP: virtual sensor modeling for fringe projection profilometry in NVIDIA Isaac Sim},
+  author={Haroon, Adam and Lakshman, Anush and Balasubramaniam, Badrinath and Li, Beiwen},
+  journal={arXiv preprint arXiv:2509.22685},
+  year={2025}
+}
+```
+
+## Contact
+
+For questions or issues, please open an issue on [GitHub](https://github.com/AnushLak/FPP-ML-Benchmarking) or contact [anushlak@iastate.edu](mailto:anushlak@iastate.edu) or [aharoon@iastate.edu](mailto:aharoon@iastate.edu).