README.md

---
license: cc-by-4.0
language:
- en
tags:
- piv
- particle-image-velocimetry
- fluid-dynamics
- turbulence
- channel-flow
- validation
- benchmark
- synthetic-data
pretty_name: PIVtools Turbulent Channel Validation Dataset
size_categories:
- 10K<n<100K
---

# PIVtools Turbulent Channel Validation Dataset

Synthetic particle-image-velocimetry (PIV) images of a turbulent channel flow at Re_τ ≈ 1000, paired with DNS-derived ground-truth statistics. Designed to benchmark PIV algorithms end-to-end against a reference dataset of known answer.

Two cases are provided:

- **Case A (clean)**: 85 000 particles per 2048 × 2048 image (≈ 5.2 ppw at 16 × 16 windows), no noise. Sets an upper bound on PIV accuracy.
- **Case B (noisy)**: 22 000 particles per image (≈ 1.3 ppw), Gaussian sensor noise (mean 80, std 16, SNR ≈ 8). Realistic experimental conditions.

Each case contains 4 000 image pairs in both planar and stereo geometries. The stereo cameras are placed symmetrically at ± 45° from the sheet normal, in a side-scatter arrangement. Ground truth is provided separately for each case — both derive from the same underlying JHTDB channel snapshots but with their respective particle counts, so finite-sample statistics are self-consistent.

Companion to the PIVtools software paper (SoftwareX, submitted). The dataset is self-contained: drop it next to a [PIVtools](https://github.com/MTT69/python-PIVtools) install and the benchmark scripts reproduce every validation figure in the paper.

## Quickstart

```bash
# 1. Install pivtools (C extensions are pre-built in the PyPI wheel)
pip install pivtools

# 2. Download this dataset
hf download MTT69/TurbulentChannel --repo-type dataset --local-dir ./tc

# 3. Process the planar noisy images (ensemble PIV — example)
pivtools-cli init --output ./work/config.yaml
# edit config.yaml to point sources at ./tc/planar_noisy
pivtools-cli ensemble --config ./work/config.yaml

# 4. Benchmark against DNS  (use ground_truth/clean for Case A, ground_truth/noisy for Case B)
python ./tc/scripts/benchmark_comparison.py \
    --mode ensemble \
    --gt-dir ./tc/ground_truth/noisy \
    --ensemble-dir ./work/calibrated_piv/4000/Cam1/ensemble \
    --num-frames 4000 \
    --output-dir ./work/validation
```

## Contents

```
MTT69/TurbulentChannel/
├── README.md                         (this file)
├── LICENSE                           (CC-BY-4.0)
├── ground_truth/
│   ├── clean/direct_stats.mat        DNS statistics for Case A (85k particles)
│   └── noisy/direct_stats.mat        DNS statistics for Case B (22k particles)
├── planar_clean/                     Case A planar images (4000 pairs, 85k particles)
│   └── B00001_A.tif … B04000_B.tif   2048 × 2048 TIFF, flat at root
├── planar_noisy/                     Case B planar images (4000 pairs, 22k particles)
│   ├── B00001_A.tif … B04000_B.tif
│   └── calibration_boards/           20 synthetic dotboard calibration images
│                                     (shared between Case A and Case B)
├── stereo_clean/                     Case A stereo images (4000 pairs × 2 cameras)
│   ├── camera1/                      cam 1 TIFFs
│   ├── camera2/                      cam 2 TIFFs
│   ├── mask_Cam1.mat                 pixel-space masks
│   └── mask_Cam2.mat
├── stereo_noisy/                     Case B stereo images (4000 pairs × 2 cameras)
│   ├── camera1/
│   ├── camera2/
│   ├── calibration/
│   │   ├── cam1/                     20 stereo dotboard images, cam 1
│   │   └── cam2/                     20 stereo dotboard images, cam 2
│   │                                 (shared between Case A and Case B)
│   ├── mask_Cam1.mat
│   └── mask_Cam2.mat
└── scripts/
    ├── benchmark_comparison.py       Planar + ensemble vs DNS
    ├── stereo_benchmark_comparison.py Stereo 3-component + 6 stresses vs DNS
    ├── cross_method_comparison.py    Multi-method overlay figures
    ├── paper_figures.py              Combined clean + noisy paper figures
    ├── tcf_direct_stats.py           Recompute ground truth from JHTDB particles
    └── sig_configs/                  EUROSIG configuration files (.cdl)
```

**Calibration boards are shared.** To avoid duplicate uploads, the calibration images for both Case A and Case B live at `planar_noisy/calibration_boards/` (planar) and `stereo_noisy/calibration/{cam1,cam2}/` (stereo). When processing Case A, point your PIVtools config at those same calibration paths.

## Image specifications

| Parameter | Value |
|-----------|-------|
| Image size | 2048 × 2048 px, 16-bit TIFF |
| Particle diameter | 3 px |
| Laser sheet thickness | 16 px (1.2 mm physical) |
| Number of pairs | 4000 |
| Case A particle count | 85 000 per image (≈ 5.2 ppw at 16 × 16 windows), no noise |
| Case B particle count | 22 000 per image (≈ 1.3 ppw at 16 × 16 windows) |
| Case B noise | Gaussian, mean = 80, std = 16, SNR ≈ 8 |
| Stereo geometry | Two cameras at ± 45° from the sheet normal (side-scatter arrangement) |
| dt | Matches JHTDB snapshot spacing (see CDL configs) |

## Ground truth

Two ground-truth files are provided, one per case, each in its own subdirectory so the benchmark scripts can point at them directly via `--gt-dir`:

- `ground_truth/clean/direct_stats.mat` — Case A reference (85 000 particle trajectories)
- `ground_truth/noisy/direct_stats.mat` — Case B reference (22 000 trajectories)

Both are computed directly from the JHTDB particle position snapshots used to render the corresponding images. Benchmark Case A PIV against the clean file and Case B against the noisy one — finite-sample statistics are self-consistent within each case. Both share this schema:

| Key | Shape | Description |
|-----|-------|-------------|
| `y_plus` | (N,) | wall-normal coordinate, wall units |
| `U_plus` | (N, 3) | mean velocity [U, V, W] in wall units |
| `stress_plus` | (N, 3, 3) | Reynolds stress tensor in wall units |
| `stress_ci_lo`, `stress_ci_hi` | (N, 3, 3) | 95% confidence interval |
| `umean_ci_lo`, `umean_ci_hi` | (N, 3) | 95% CI for mean velocity |
| `u_tau` | scalar | friction velocity (mm/s) |
| `delta_nu` | scalar | viscous length scale (mm) |
| `Re_tau` | scalar | friction Reynolds number |

The Case B ground truth is self-consistent with the 22 000-particle rendering — finite-sample statistics from the subsampled particle set, not the full DNS. Benchmark Case B PIV against Case B ground truth.

## How this was generated

Synthetic images are rendered from JHTDB turbulent-channel particle trajectories using the **EUROSIG / EUROPIV synthetic image generator**. The configuration files in `scripts/sig_configs/` are the authoritative build instructions:

| Configuration | Role |
|---------------|------|
| `sigconf_planar_noisy_A.cdl` | Planar frame A, 22k particles, noise pattern A |
| `sigconf_planar_noisy_B.cdl` | Planar frame B, 22k particles, noise pattern B |
| `SIGconf_Stereo_cam1_noisy_A.cdl`, `..._B.cdl` | Stereo cam 1, frames A and B |
| `SIGconf_Stereo_cam2_noisy_A.cdl`, `..._B.cdl` | Stereo cam 2, frames A and B |
| `sigconf_planar.cdl`, `SIGconf_Stereo_cam{1,2}.cdl` | Case A planar + stereo (85k particles, no noise) |

To regenerate images bit-for-bit, install EUROSIG and invoke each `.cdl` with its associated particle-position files from JHTDB. See the SIG documentation for build instructions.

## Scripts

### `benchmark_comparison.py` — single-method benchmark

Compares planar or ensemble PIV against the DNS ground truth; produces U+, Reynolds stress, residual, trace invariant, and noise-decomposition plots.

```bash
python scripts/benchmark_comparison.py \
    --mode ensemble \
    --gt-dir ./ground_truth/noisy \
    --ensemble-dir <path/to/your/ensemble_result_directory> \
    --num-frames 4000 \
    --output-dir ./out
```

| Flag | Description |
|------|-------------|
| `--mode` / `-m` | `instantaneous` or `ensemble` |
| `--runs` / `-r` | Comma-separated 0-based pass indices (e.g. `2,3`) |
| `--windows` / `-w` | Labels for those passes (e.g. `32,16`) |
| `--gt-dir` / `-g` | Directory containing `direct_stats.mat` — e.g. `./ground_truth/noisy` or `./ground_truth/clean` (required) |
| `--base-dir` / `-b` | PIV results base (instantaneous mode) |
| `--ensemble-dir` / `-e` | Direct path to ensemble result directory |
| `--num-frames` / `-n` | Frame count subdirectory (default 1000; use 4000 for this dataset) |
| `--output-dir` / `-o` | Output directory |
| `--y-plus-offset` / `-y` | Additional y+ offset on top of hardcoded +1 |
| `--show-fit-lines` | Overlay log-law and viscous sublayer curves |

### `stereo_benchmark_comparison.py` — stereo 3C + 6-stress benchmark

Uses LaTeX for labels (`text.usetex=True`); requires MiKTeX / TeXLive.

```bash
python scripts/stereo_benchmark_comparison.py \
    --gt-dir ./ground_truth/noisy \
    --stereo-base <path/to/your/stereo_results> \
    --num-frames 4000 \
    --output-dir ./out
```

### `cross_method_comparison.py` — multi-method overlay

Publication-quality plots comparing one pass from each of instantaneous, ensemble, and stereo against DNS on the same axes. Okabe-Ito colourblind palette.

```bash
python scripts/cross_method_comparison.py \
    --gt-dir ./ground_truth/noisy \
    --output-dir ./out \
    --inst-stats <path/to/instantaneous/mean_stats.mat> \
    --ens-dir <path/to/ensemble_dir> \
    --stereo-stats <path/to/stereo/mean_stats.mat>
```

### `paper_figures.py` — combined Case A + Case B figures

Reproduces the figures in the PIVtools paper: Case A (open symbols) and Case B (filled symbols) overlaid. Any combination of paths may be supplied — the script plots whichever it receives.

```bash
# Case B only (what this dataset ships today)
python scripts/paper_figures.py \
    --gt-noisy-dir ./ground_truth \
    --inst-noisy-stats <path/to/noisy/instantaneous/mean_stats.mat> \
    --ens-noisy-dir <path/to/noisy/ensemble_dir> \
    --stereo-noisy-stats <path/to/noisy/stereo/mean_stats.mat> \
    --output-dir ./out
```

### `tcf_direct_stats.py` — recompute ground truth

If you regenerate the synthetic images via EUROSIG, this script recomputes `direct_stats.mat` from the underlying JHTDB particle position files (`B*_A.data`, `B*_B.data`).

```bash
python scripts/tcf_direct_stats.py \
    --data-dir <path/to/particle_positions> \
    --output-dir ./ground_truth
```

## Unit conventions

| Quantity | PIVtools storage | Benchmark display |
|----------|-----------------|-------------------|
| Velocity | m/s | mm/s (× 1000) |
| Reynolds stress | (m/s)² | (mm/s)² (× 1e6) |
| Spatial coordinates | mm | wall units y⁺ = y / δ_ν |

## Masks

`stereo_noisy/mask_Cam{1,2}.mat` hold pixel-space boolean masks (same shape as images) that exclude regions outside the valid field of view. PIVtools loads them automatically when configured with `masking.enabled: true` and `mask_file_pattern: mask_Cam{cam}.mat`.

## Citation

If you use this dataset, please cite both the PIVtools paper and the underlying DNS source.

```bibtex
@article{taylor_pivtools,
  title={PIVtools: an open-source PIV framework with integrated planar, stereoscopic, and ensemble pipelines},
  author={Taylor, M.T. and Lawson, J.M. and Ganapathisubramani, B.},
  journal={SoftwareX},
  note={submitted}
}

@article{lee2015direct,
  title={Direct numerical simulation of turbulent channel flow up to Re_tau = 5200},
  author={Lee, M. and Moser, R.D.},
  journal={J. Fluid Mech.},
  year={2015}
}

@article{li2008public,
  title={A public turbulence database cluster and applications to study Lagrangian evolution of velocity increments in turbulence},
  author={Li, Y. and others},
  journal={J. Turbulence},
  year={2008}
}
```

DNS reference data is from the **Johns Hopkins Turbulence Database** (JHTDB).

## License

CC-BY-4.0 — free to use, modify, and redistribute with attribution to the PIVtools paper.

The DNS ground truth is derived from publicly accessible JHTDB data and is redistributed here under the same permissive terms; consult the JHTDB usage policy (http://turbulence.pha.jhu.edu) for their citation requirements.