eturok-weizmann/laser-vibrations

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; as loudspeakers excite the box, each laser's speckle pattern shifts in proportion to the local surface vibration. The goal is to reconstruct the shape and location of an object inside the box from the vibration signals alone.

Per-sample viewer metadata lives in data/metadata.jsonl. Full signal data and media files live in per-sample subdirectories under data/. Shared overhead images live under image/ and the shared audio file under audio/.

---

Dataset Columns

These are the columns shown in the HuggingFace dataset viewer, sourced from data/metadata.jsonl.

Column	Type	Description
sample_id	int	Unique sequential identifier for this sample
overhead_file_name	image	Final per-sample overhead photo shown in the viewer; stored as data/<sample_id>/overhead.png and includes the speaker overlay
speckle_vibrations_file_name	video	Slow-motion preview of the laser speckle pattern while the box vibrates
speckle_shifts_ifft_audio_file_name	audio	Vibration signal of a single laser point reconstructed as audio (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 means only speaker 4 was 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)
object	string	Object type inside the box, e.g. cube
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 Section 3)

---

File Directory Structure

experiment-16/
├── README.md                            # This file (dataset card)
├── audio/
│   └── chirp_50_1000_3.0sec.wav         # Shared excitation chirp (50–1000 Hz, 3 s)
├── data/
│   ├── metadata.jsonl                   # One JSON row per sample (viewer-facing)
│   ├── 0000001/                         # Per-sample directory (zero-padded 7-digit ID)
│   │   ├── manifest.json                # Full provenance + config for this sample
│   │   ├── overhead.png                 # Final overhead image with padding + speaker overlay 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       # Slow-motion preview video of speckle motion
│   └── 0000002/
│       └── ...
└── image/
    └── <image_dir>/                     # Named: <object>-<x>x-<y>y-<n>obj-<material>-<date>
        ├── raw_overhead.png             # Full overhead photo before cropping
        ├── cropped_overhead.png         # Overhead cropped to the box region
        ├── segmented_overhead.png       # Shared overhead with mask overlay + COM marker only (no speaker overlay)
        ├── mask.png                     # Binary segmentation mask (white = object)
        └── mask.npz                     # Binary mask as compressed numpy array

---

manifest.json Structure

Every sample directory contains a manifest.json that records full provenance, hardware config, processing parameters, and artifact paths. The manifest column in metadata.jsonl is this same document serialised as a JSON string. Shared image assets live under image/<image_dir>/, while the sample-specific speaker-rendered overhead lives under data/<sample_id>/overhead.png.

Top-Level Keys

The most important, top-level entries of manifest.json.

Key	Type	Description
sample_id	int	Unique sample identifier
experiment_id	string	Source recording directory name (from experiment-15)
experiment_dir	string	Target directory name (experiment-16)
source_experiment_id	string	Canonical source reference (same as experiment_id)
source_experiment_dir	string	Absolute NAS path to the source directory
hf_repo	string	HuggingFace repo this sample was uploaded to
sample	object	Physical setup at the time of recording
segmentation	object	Overhead-image segmentation result
experiment_config	object	Merged hardware and recording configuration
experiment_output	object	Derived statistics computed during processing
processing_config	object	Processing pipeline parameters
artifacts	object	Relative repo paths to all files produced for this sample, including both shared and sample-specific overhead images

sample

The basic attributes of a single sample of our dataset.

Key	Type	Description
object	string	Object type inside the box, e.g. cube or empty
n_objects	int	Number of objects
box_material	string	Box material, e.g. cardboard
speakers	string	4-bit activation code for active speakers, e.g. 0001
x_position	int	Object grid column (0-indexed)
y_position	int	Object grid row (0-indexed)
image_dir	string	Image subdirectory name under data/image/

segmentation

Values from the segmentation mask.

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

experiment_config

Processing parameters derived (and possibly overidden) from base_experiment_config.json

Key	Type	Description
audio.file_name	string	Path to the excitation audio file within the repo
audio.sample_rate_hz	int	Audio sample rate (Hz)
audio.duration_s	float	Total audio duration including silence padding (s)
audio.total_output_channels	int	Number of output channels on the audio interface
audio.wav_channels	int	Number of channels in the WAV file
audio.sample_width_bytes	int	Sample width in bytes
audio.generation.signal	string	Signal type, e.g. chirp
audio.generation.method	string	Chirp sweep method, e.g. logarithmic
audio.generation.chirp_duration_s	float	Duration of the chirp sweep (s)
audio.generation.silence_start_s	float	Silence padding before the chirp (s)
audio.generation.silence_end_s	float	Silence padding after the chirp (s)
audio.generation.f_start_hz	int	Chirp start frequency (Hz)
audio.generation.f_end_hz	int	Chirp end frequency (Hz)
audio.generation.output_dtype	string	Output sample dtype, e.g. int16
audio.generation.normalization	string	Normalization method, e.g. peak_to_int16
recording.capture_seconds_requested	float	Requested laser camera recording duration (s)
overhead_camera.frame_rate_fps	int	Overhead camera frame rate (fps)
overhead_camera.exposure_ms	int	Overhead camera exposure time (ms)
overhead_camera.pixel_clock_mhz	int	Overhead camera pixel clock (MHz)
overhead_camera.gain	int	Overhead camera sensor gain
overhead_camera.runtime.device_id	int	Camera device index
overhead_camera.runtime.color_mode	string	Raw sensor color mode
overhead_camera.runtime.buffer_count	int	Frame buffer count
overhead_camera.runtime.rotation_degrees	int	Image rotation applied at capture (degrees)
overhead_camera.runtime.debayer_code	string	OpenCV debayer conversion code
laser_camera.runtime.info_field	bool	Whether the camera info field overlay is enabled
laser_camera.runtime.cxp_link_configuration	string	CoaXPress link configuration
laser_camera.calibration.fps	int	Frame rate used during ROI calibration (fps)
laser_camera.calibration.exposure_us	int	Exposure time used during calibration (µs)
laser_camera.calibration.gain	int	Gain used during calibration
laser_camera.capture.fps	int	Frame rate used during recording (fps)
laser_camera.capture.exposure_us	int	Exposure time used during recording (µs)
laser_camera.capture.gain	int	Gain used during recording
laser_camera.capture.buffer_part_count	int	Number of frame buffer partitions
laser_grid.n_roi_rows	int	Number of laser rows in the grid
laser_grid.n_roi_columns	int	Number of laser columns in the grid
laser_grid.roi_row_height	int	Height of each laser ROI (pixels)
laser_grid.roi_column_width	int	Width of each laser ROI (pixels)
preview.overhead_resize_factor	float	Resize factor for the overhead preview display
preview.overhead_gamma	float	Gamma correction for the overhead preview
preview.laser_preview_gamma	float	Gamma correction for the laser preview
preview.show_full_frame	int	Whether to display the full camera frame (0/1)
preview.preview_level	int	Preview detail level
preview.reset_rois	bool	Whether to reset ROIs on startup

experiment_output

The output of the data collection phase.

Key	Type	Description
overhead_camera.image_width	int	Overhead image width (px)
overhead_camera.image_height	int	Overhead image height (px)
laser_camera.global_roi	list[int]	Full-frame ROI used [x, y, w, h]
laser_camera.max_frame_rate_hz	int	Maximum frame rate achieved by the laser camera
laser_grid.total_image_height	int	Total laser camera frame height (px)
laser_grid.selected_row_points_image_xy	list[list[int]]	Per-row representative laser point coordinates [[x, y], ...] in the camera frame
laser_grid.selected_column_centers_x	list[int]	X-centre pixel of each laser column in the cropped frame
laser_grid.row_values_single_list	list[int]	Row pixel indices belonging to all laser ROIs (flattened)
laser_grid.global_crop_x	int	X offset of the global crop in the full camera frame (px)
laser_grid.global_crop_width	int	Width of the global crop (px)
laser_grid.global_crop_height	int	Height of the global crop (px)
laser_grid.row_rois_y	list[list[int]]	Y extents of each laser row [[y_start, y_end], ...] in the full frame
laser_grid.sensor_grid_shape	list[int]	Laser grid dimensions [rows, cols]
laser_grid.sensor_rois_xywh	list[list[int]]	Per-laser bounding boxes [[x, y, w, h], ...] within the cropped frame
speckle_vibrations.frame_count	int	Total number of frames captured
speckle_vibrations.frame_height	int	Raw frame height (px)
speckle_vibrations.frame_width	int	Raw frame width (px)
speckle_vibrations.preview_frame_count	int	Number of frames in the preview video
speckle_vibrations.preview_fps	float	Playback frame rate of the preview video (fps)
speckle_vibrations.preview_width	int	Preview video width (px)
speckle_vibrations.preview_height	int	Preview video height (px)
speckle_vibrations.capture_fps_hz	float	Actual laser camera capture rate (Hz)
speckle_vibrations.capture_seconds_requested	float	Requested recording duration (s)
speckle_vibrations.capture_seconds_observed	float	Observed recording duration (s)
speckle_vibrations.dtype	string	Raw frame pixel dtype
speckle_shifts.fs	float	Shift signal sampling rate (Hz)
speckle_shifts.shape	list[int]	Shape of the shifts array [n_lasers, n_frames, 2]
speckle_shifts_clean.shape	list[int]	Shape of the filtered shifts array [n_lasers, n_frames, 2]
speckle_shifts_clean.fs	float	Sampling rate after filtering (Hz)
speckle_shifts_fft.shape	list[int]	Shape of the FFT array [n_lasers, n_freqs, 2]
speckle_shifts_fft.fs	float	Sampling rate used for the FFT (Hz)
speckle_shifts_fft.n_samples	int	Number of time-domain samples used for the FFT
speckle_shifts_ifft_audio.sample_rate_hz	int	Sample rate of the reconstructed audio WAV (Hz)
speckle_shifts_ifft_audio.n_frames	int	Number of audio samples in the output WAV

processing_config

Processing parameters derived (and possibly overidden) from base_processing_config.json.

Key	Type	Description
speckle_vibration_raw.format	string	Storage format: npy or npz
speckle_vibration_raw.compressed	bool	Whether the raw array is compressed
speckle_vibrations_preview.max_frames	int	Maximum number of frames in the preview video
speckle_vibrations_preview.max_width	int	Maximum preview video width (px)
speckle_vibrations_preview.macro_block_size	int	Encoder macro-block alignment size
speckle_vibrations_preview.source_capture_fps_hz	float	Source capture rate used for preview timing (Hz)
speckle_vibrations_preview.preserve_physical_duration	bool	Whether playback speed matches real time
speckle_vibrations_preview.percentile_low	float	Lower percentile for frame contrast normalisation
speckle_vibrations_preview.percentile_high	float	Upper percentile for frame contrast normalisation
speckle_vibrations_preview.codec	string	Video codec, e.g. libx264
speckle_vibrations_preview.pixelformat	string	Pixel format, e.g. yuv420p
speckle_vibrations_preview.crf	int	Constant rate factor — lower = higher quality
speckle_vibrations_preview.burn_frame_index	bool	Whether the frame index is burned into the video
speckle_shifts.fs_hz	float	Sampling rate of the shift signal (Hz)
speckle_shifts_clean.filter_type	string	Filter design, 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 after filtering
speckle_shifts_clean.apply_order	string	filter_then_hann or hann_then_filter
speckle_shifts_fft.fft_kind	string	FFT variant, e.g. rfft
speckle_shifts_fft.fft_axis	int	Axis along which the FFT is computed
speckle_shifts_fft.min_freq	float	Minimum frequency retained after crop (Hz)
speckle_shifts_fft.max_freq	float	Maximum frequency retained after crop (Hz)
speckle_shifts_fft.dtype	string	Complex dtype of the output, e.g. complex64
speckle_shifts_fft.crop_after_fft	bool	Whether to crop to [min_freq, max_freq] after FFT
speckle_shifts_ifft_audio.laser_idx	int	Which laser (0-indexed) to use for the audio preview
speckle_shifts_ifft_audio.xy_idx	int	Which shift channel: 0 = X, 1 = Y
speckle_shifts_ifft_audio.method	string	Reconstruction method, e.g. ifft
speckle_shifts_ifft_audio.output_sample_rate_hz	int	Output WAV sample rate (Hz)
speckle_shifts_ifft_audio.normalization	string	Normalization method, e.g. peak_to_int16
speckle_shifts_ifft_audio.output_dtype	string	Output sample dtype, e.g. int16
speckle_shifts_ifft_audio.zero_fill_uncropped_bins	bool	Whether to zero-fill bins outside the crop range
segmentation.left	float	Left crop fraction of the overhead image
segmentation.right	float	Right crop fraction of the overhead image
segmentation.up	float	Top crop fraction of the overhead image
segmentation.down	float	Bottom crop fraction of the overhead image
segmentation.prompt	string	Text prompt passed to the segmentation model

artifacts

The final artifacts we end up with in our dataset.

Key	Type	Description
raw_overhead	string	data/image/<dir>/raw_overhead.png — full overhead photo before cropping
cropped_overhead	string	data/image/<dir>/cropped_overhead.png — overhead cropped to box region
segmented_overhead	string	data/image/<dir>/segmented_overhead.png — shared overhead with mask + COM marker only
overhead	string	data/<id>/overhead.png — final per-sample overhead with padding + speaker overlay
mask_png	string	data/image/<dir>/mask.png — binary segmentation mask (PNG)
mask_npz	string	data/image/<dir>/mask.npz — binary mask as compressed numpy array
audio	string	data/audio/chirp_50_1000_3.0sec.wav — shared excitation chirp (all samples)
speckle_vibration_raw	string	data/<id>/speckle_vibration_raw.npy — raw laser camera frames
speckle_vibrations	string	data/<id>/speckle_vibrations.mp4 — slow-motion speckle preview video
speckle_shifts	string	data/<id>/speckle_shifts.npz — sub-pixel XY shifts per laser per frame
speckle_shifts_clean	string	data/<id>/speckle_shifts_clean.npz — filtered + windowed shifts
speckle_shifts_fft	string	data/<id>/speckle_shifts_fft.npz — FFT of cleaned shifts
speckle_shifts_ifft_audio	string	data/<id>/speckle_shifts_ifft_audio.wav — single-laser audio reconstruction
manifest	string	data/<id>/manifest.json — this manifest file