---
configs:
  - config_name: default
    data_files:
      - data/metadata.jsonl
dataset_info:
  features:
    - name: sample_id
      dtype: int64
    - name: segmented_overhead_file_name
      dtype: image
    - name: speckle_vibrations_file_name
      dtype: video
    - name: speckle_shifts_ifft_audio_file_name
      dtype: audio
    - name: audio_file_name
      dtype: audio
    - name: experiment_id
      dtype: string
    - name: speakers
      dtype: string
    - name: x_position
      dtype: int64
    - name: y_position
      dtype: int64
    - name: x_com
      dtype: float64
    - name: y_com
      dtype: float64
    - name: n_objects
      dtype: int64
    - name: box_material
      dtype: string
    - name: mask_file_name
      dtype: image
    - name: experiment_dir
      dtype: string
    - name: manifest
      dtype: string
---

# Laser Vibrations

Dataset of laser speckle vibration recordings used to locate objects hidden inside a cardboard box.
A 10×10 grid of lasers shines on the side of a box containing an object; as loudspeakers excite the box,
the speckle patterns shift in proportion to the local surface vibration. Per-sample metadata is in
`data/metadata.jsonl`; full signal data and media files live in per-sample subdirectories.

## Dataset Viewer Columns

| Column | Type | Description |
|--------|------|-------------|
| `sample_id` | int | Unique sequential identifier for this sample |
| `segmented_overhead_file_name` | image | Overhead photo of the box with segmentation mask overlay and speaker-angle annotations |
| `speckle_vibrations_file_name` | video | Slow-motion preview video of the laser speckle pattern on the box side while vibrating |
| `speckle_shifts_ifft_audio_file_name` | audio | Reconstructed audio from the vibration signal of a single laser point (inverse FFT) |
| `audio_file_name` | audio | Shared excitation chirp played through the loudspeakers during recording |
| `experiment_id` | string | Source directory name from experiment-15 (unique per recording) |
| `speakers` | string | 4-bit speaker activation code, e.g. `0001` = only speaker 4 active |
| `x_position` | int | Object grid column index (0-indexed) |
| `y_position` | int | Object grid row index (0-indexed) |
| `x_com` | float | X centre-of-mass of the segmentation mask in the cropped overhead image (pixels) |
| `y_com` | float | Y centre-of-mass of the segmentation mask in the cropped overhead image (pixels) |
| `n_objects` | int | Number of objects inside the box |
| `box_material` | string | Box material, e.g. `cardboard` |
| `mask_file_name` | image | Binary segmentation mask of the object in the cropped overhead image |
| `experiment_dir` | string | Name of the experiment-16 target directory |
| `manifest` | string | Full JSON manifest for this sample (see Manifest section below) |

## File Directory Structure

```
experiment-16/
├── README.md                        # This file (dataset card)
├── data/
│   ├── metadata.jsonl               # One JSON row per sample (viewer-facing)
│   ├── audio/
│   │   └── chirp_50_1000_3.0sec.wav # Shared excitation signal played by all speakers
│   ├── 0000001/                     # Per-sample data directory (zero-padded 7-digit ID)
│   │   ├── manifest.json            # Full metadata + config for this sample
│   │   ├── speckle_vibration_raw.npy        # Raw laser camera frames (100 lasers × T frames × 2 XY)
│   │   ├── speckle_shifts.npz               # Sub-pixel XY shifts per laser per frame
│   │   ├── speckle_shifts_clean.npz          # Bandpass-filtered + Hann-windowed shifts
│   │   ├── speckle_shifts_fft.npz            # FFT of cleaned shifts (frequency domain)
│   │   ├── speckle_shifts_ifft_audio.wav     # Single-laser vibration reconstructed as audio
│   │   └── speckle_vibrations.mp4            # Preview video of speckle motion
│   ├── 0000002/
│   │   └── ...
│   └── image/
│       └── <image_dir>/             # Per-sample image directory (named by object+position+date)
│           ├── raw_overhead.png     # Raw overhead photo before cropping
│           ├── cropped_overhead.png # Overhead photo cropped to box region
│           ├── segmented_overhead.png   # Overhead with mask overlay + speaker annotations
│           ├── mask.png             # Binary mask (white = object)
│           └── mask.npz             # Binary mask as compressed numpy array
```

## manifest.json Structure

Each sample directory contains a `manifest.json` with full provenance and configuration.
The `manifest` column in `metadata.jsonl` is this same JSON serialised as a string.

### Top-level keys

| Key | Type | Description |
|-----|------|-------------|
| `sample_id` | int | Unique sample identifier |
| `experiment_id` | string | Source recording directory name |
| `experiment_dir` | string | Target experiment-16 directory name |
| `source_experiment_id` | string | Same as `experiment_id` (canonical source reference) |
| `source_experiment_dir` | string | Absolute NAS path to source directory |
| `hf_repo` | string | HuggingFace repo this sample was uploaded to |
| `sample` | object | Physical setup description — see below |
| `segmentation` | object | Segmentation result — see below |
| `experiment_config` | object | Merged hardware + recording config — see below |
| `experiment_output` | object | Derived signal statistics from processing — see below |
| `processing_config` | object | Processing pipeline parameters — see below |
| `artifacts` | object | Relative paths to all files for this sample — see below |

### `sample`

| Key | Type | Description |
|-----|------|-------------|
| `object` | string | Object type inside the box, e.g. `cube` |
| `n_objects` | int | Number of objects |
| `box_material` | string | Box material, e.g. `cardboard` |
| `speakers` | string | 4-bit activation code for which speakers were active |
| `x_position` | int | Object grid column (0-indexed) |
| `y_position` | int | Object grid row (0-indexed) |
| `image_dir` | string | Name of the image subdirectory under `data/image/` |

### `segmentation`

| Key | Type | Description |
|-----|------|-------------|
| `x_com` | float | X centre-of-mass of segmentation mask (pixels in cropped image) |
| `y_com` | float | Y centre-of-mass of segmentation mask (pixels in cropped image) |
| `status` | string | `completed` if segmentation succeeded |

### `experiment_config`

Merged from the source `experiment_config.json` and hardware defaults.

| Key | Type | Description |
|-----|------|-------------|
| `audio.file_name` | string | Audio file name used for excitation |
| `audio.sample_rate_hz` | int | Audio sample rate in Hz |
| `audio.duration_s` | float | Duration of the chirp in seconds |
| `audio.generation.f_start_hz` | int | Chirp start frequency (Hz) |
| `audio.generation.f_end_hz` | int | Chirp end frequency (Hz) |
| `recording.capture_seconds_requested` | float | Requested recording duration (s) |
| `overhead_camera.frame_rate_fps` | int | Overhead camera frame rate |
| `laser_camera.capture.fps` | float | Laser camera frame rate |
| `laser_camera.global_roi` | list[int] | Full-frame ROI [x, y, w, h] |
| `laser_grid.sensor_rois_xywh` | list[list[int]] | Per-laser ROI bounding boxes |

### `experiment_output`

Derived statistics computed during processing.

| Key | Type | Description |
|-----|------|-------------|
| `overhead_camera.image_width` | int | Overhead image width (px) |
| `overhead_camera.image_height` | int | Overhead image height (px) |
| `laser_camera.max_frame_rate_hz` | int | Actual laser camera frame rate achieved |
| `laser_camera.global_roi` | list[int] | Actual ROI used |
| `laser_grid.total_image_height` | int | Total height of the laser camera frame |
| `laser_grid.n_lasers` | int | Number of laser points detected |
| `speckle_shifts.n_frames` | int | Number of frames captured |
| `speckle_shifts.duration_s` | float | Actual recording duration (s) |

### `processing_config`

Parameters used for each processing stage.

| Key | Type | Description |
|-----|------|-------------|
| `speckle_vibration_raw.format` | string | Raw data format: `npy` or `npz` |
| `speckle_vibration_raw.compressed` | bool | Whether raw data is compressed |
| `speckle_shifts.fs_hz` | float | Sampling rate of shift signal (Hz) |
| `speckle_shifts_clean.filter_type` | string | Filter type, e.g. `butterworth` |
| `speckle_shifts_clean.filter_mode` | string | `bandpass`, `lowpass`, or `highpass` |
| `speckle_shifts_clean.lowcut` | float | Low cutoff frequency (Hz) |
| `speckle_shifts_clean.highcut` | float | High cutoff frequency (Hz) |
| `speckle_shifts_clean.filter_order` | int | Filter order |
| `speckle_shifts_clean.hann_applied` | bool | Whether a Hann window was applied |
| `speckle_shifts_fft.fft_kind` | string | FFT variant, e.g. `rfft` |
| `speckle_shifts_fft.min_freq` | float | Minimum frequency kept after crop (Hz) |
| `speckle_shifts_fft.max_freq` | float | Maximum frequency kept after crop (Hz) |
| `speckle_shifts_fft.dtype` | string | Complex dtype, e.g. `complex64` |
| `speckle_shifts_ifft_audio.laser_idx` | int | Which laser index to use for audio preview |
| `speckle_shifts_ifft_audio.output_sample_rate_hz` | int | Audio output sample rate (Hz) |
| `speckle_shifts_ifft_audio.normalization` | string | Normalization method for audio |
| `segmentation.left/right/up/down` | float | Crop fractions used for overhead image |
| `segmentation.prompt` | string | Text prompt passed to the segmentation model |
| `speckle_vibrations_preview.codec` | string | Video codec, e.g. `libx264` |
| `speckle_vibrations_preview.crf` | int | Video quality (lower = better) |
| `speckle_vibrations_preview.max_frames` | int | Max frames in preview video |

### `artifacts`

Relative paths (within the repo) to every file produced for this sample.

| Key | Description |
|-----|-------------|
| `raw_overhead` | Raw overhead photo before cropping |
| `cropped_overhead` | Overhead photo cropped to box region |
| `segmented_overhead` | Overhead with mask overlay and speaker annotations |
| `mask_png` | Binary segmentation mask as PNG |
| `mask_npz` | Binary segmentation mask as compressed numpy |
| `audio` | Shared excitation chirp WAV |
| `speckle_vibration_raw` | Raw laser camera shift data (npy/npz) |
| `speckle_vibrations` | Preview video of speckle motion (mp4) |
| `speckle_shifts` | Sub-pixel XY shifts per laser per frame (npz) |
| `speckle_shifts_clean` | Bandpass-filtered + Hann-windowed shifts (npz) |
| `speckle_shifts_fft` | FFT of cleaned shifts in frequency domain (npz) |
| `speckle_shifts_ifft_audio` | Single-laser vibration as audio WAV |
| `manifest` | This manifest.json file |