File Formats - UXO Dataset 2024

The UXO Dataset 2024 uses simple, well-established file formats for maximum accessibility and ease of use.

Image Formats

PGM - ARIS Raw Frames

ARIS raw frames are stored as 8-bit grayscale PGM (Portable Gray Map) images. Format: .pgm Characteristics:

Grayscale images representing raw sonar beam intensities
Each pixel represents a single sample from a beam
Beams are ordered right to left in the image
Dimensions vary based on ping mode (48, 64, 96, or 128 beams)

The raw ARIS frames store beam data ordered right to left. When displaying, you may want to flip the image for intuitive viewing (see view_recording.py:180).

Beam Count by Ping Mode:

def get_beamcount_from_pingmode(pingmode):
    if pingmode in [1, 2]:
        return 48
    if pingmode in [3, 4, 5]:
        return 96
    if pingmode in [6, 7, 8]:
        return 64
    if pingmode in [9, 10, 11, 12]:
        return 128

From aris_definitions.py:4-17

PNG - ARIS Polar Frames

ARIS polar frames represent the sonar data transformed into the actual field-of-view geometry. Format: .png Characteristics:

Polar coordinate transformation of raw beam data
Represents what the sonar actually “sees”
Can be colorized for visualization (e.g., using TWILIGHT_SHIFTED colormap)
Better for human interpretation and visualization

Polar Transformation: The polar transformation accounts for:

Beam angles (see BeamWidthsAris3000_64 and BeamWidthsAris3000_128 in aris_definitions.py)
Range bins along each beam
Fan-shaped acoustic field of view (±15° horizontal)

ARIS 3000 Beam Angles (64 beams)

The ARIS 3000 in 64-beam mode has the following angular coverage:

Horizontal FOV: ±15°
Center beam angles: Range from -14.89° to +14.89°
Beam width: ~0.46° per beam

Full beam angle definitions are provided in aris_definitions.py:251-316 as triplets of (center, left, right) angles for each beam.

ARIS 3000 Beam Angles (128 beams)

The ARIS 3000 in 128-beam mode has:

Horizontal FOV: ±15°
Center beam angles: Range from -15.01° to +15.02°
Beam width: ~0.24° per beam

Full beam angle definitions are provided in aris_definitions.py:319-448.

JPG - GoPro Camera Frames

Camera frames from the GoPro Hero 8. Format: .jpg Characteristics:

Downsampled from original UHD resolution
Default export resolution: FHD (1920×1080)
Synchronized to ARIS frame timestamps
Compressed JPEG format for reduced file size

Labels were created on UHD resolution frames. If using FHD or other downsampled resolutions, multiply label coordinates by 3.

Tabular Data Formats

CSV - Metadata and Poses

Several CSV files store synchronized metadata and pose information.

aris_frame_meta.csv

Contains per-frame metadata from the ARIS sonar. Schema (selected fields from 240+ total):

Field	Type	Description
`FrameIndex`	int	Frame number (primary key)
`FrameTime`	uint64	Frame timestamp in microseconds
`SonarPan`	float	Sonar pan angle (degrees)
`SonarTilt`	float	Sonar tilt angle (degrees)
`SonarRoll`	float	Sonar roll angle (degrees)
`PingMode`	int	ARIS ping mode (determines beam count)
`SamplesPerBeam`	int	Number of range samples per beam
`SoundSpeed`	float	Speed of sound in water (m/s)
`WindowStart`	float	Range window start (meters)
`WindowLength`	float	Range window length (meters)
`Salinity`	float	Water salinity (ppt)
`WaterTemp`	float	Water temperature (°C)

Full Frame Header Fields

The complete list of 240+ ARIS frame header fields is defined in aris_definitions.py:69-247. Key categories include:

Timing: FrameIndex, FrameTime, timestamps
Sonar pose: SonarX, SonarY, SonarZ, SonarPan, SonarTilt, SonarRoll
Acquisition: PingMode, SamplesPerBeam, SampleRate, SoundSpeed
Environment: WaterTemp, Salinity, Pressure
Status: Battery, various fault flags
Navigation: GPS data, heading, depth, altitude

gantry.csv

Gantry crane position data interpolated to ARIS frame times. Schema:

aris_frame_idx,x,y,z
0,1.234567,2.345678,3.456789
1,1.234568,2.345679,3.456790
...

Field	Type	Description
`aris_frame_idx`	int	Corresponding ARIS frame index
`x`	float	Gantry X position (meters)
`y`	float	Gantry Y position (meters)
`z`	float	Gantry Z position (meters)

Positions are in the world coordinate frame.

ar3.csv

AR3 robot arm poses calculated from gantry position and ARIS metadata. Schema:

aris_frame_idx,pos.x,pos.y,pos.z,rot.x,rot.y,rot.z,rot.w
0,1.234567,2.345678,3.456789,0.0,0.0,0.0,1.0
...

Field	Type	Description
`aris_frame_idx`	int	Corresponding ARIS frame index
`pos.x`, `pos.y`, `pos.z`	float	AR3 position (meters)
`rot.x`, `rot.y`, `rot.z`, `rot.w`	float	AR3 orientation (quaternion)

The AR3 pose combines:

Gantry position
Gantry-to-AR3 transform from calibration/transforms.yaml
ARIS roll/tilt/pan angles from frame metadata

See release_1_export.py:18-36 for the calculation.

YAML - Calibration and Metadata

aris_file_meta.yaml

File-level ARIS recording metadata. Format: YAML key-value pairs Example fields:

Version: 5
FrameCount: 1234
FrameRate: 5.0
HighResolution: 1
NumRawBeams: 128
SampleRate: 2.0
SamplesPerChannel: 1024
ReceiverGain: 40
SN: 12345
strDate: "2024-06-15"

Full file header definition in aris_definitions.py:26-67.

transforms.yaml

Calibration transforms between all coordinate frames. Format: Hierarchical YAML with transform definitions Structure:

setup:
  - frame_id: "setup/ar3"
    parent_frame_id: "setup/portal_crane"
    translation:
      x: 0.0
      y: 0.174
      z: -0.338
    rotation:  # Quaternion (x, y, z, w)
      x: 0.0
      y: 1.0
      z: 0.0
      w: 0.0
  - frame_id: "setup/aris"
    parent_frame_id: "setup/ar3"
    ...

targets:
  - frame_id: "target/100lbs"
    parent_frame_id: "world"
    translation:
      x: 1.251079
      y: 1.592704
      z: -1.489828
    rotation:
      x: 0.0
      y: 0.0
      z: 0.0
      w: 1.0

See Target Details for more information.

Text Formats

notes.txt

Free-form notes about each recording. Format: Plain text Contents:

Target type identifier (e.g., “Target: 100lbs”)
Recording conditions
Special notes or observations
Trajectory description

Example:

Target: 100lbs
Trajectory: Forward pass, 2m distance
Conditions: Clear water, good visibility
Notes: Multiple views of the same target

The target type line is parsed by the export script to organize recordings into folders (release_1_export.py:38-43).

Binary Formats

ARIS Frame Data Structure

While the exported dataset uses standard image formats, the original ARIS recordings use a binary format with specific header structures. File Header: 1024 bytes (defined by FileHeaderDefinition) Frame Header: 1024 bytes per frame (defined by FrameHeaderDefinition) Frame Data: Variable size based on beam count and samples per beam From aris_definitions.py:21-23:

FileHeaderDefinition = '< IIIII f II ff II 32s 256s iiii IIIIIIIIIIIIIIIIII QQQ II 568x'
FrameHeaderDefinition = '< I Q II Q IIIIII ff I i IIIIII ffffffffffffffffffff dd f I fff IIII f I fffffffff d f III fff IIIIIII ffffff 16f ffff IIIIII ff IIIIII Q IIIIII ffff IIIIIIIIIIIIIIIII fff III ffffffffffff IIIIII HH Q fff II 288x'

These binary structures are parsed during preprocessing and exported to the simpler CSV and image formats.

Format Summary Table

Extension	Type	Usage	Location
`.pgm`	Image	ARIS raw frames	`aris_raw/`
`.png`	Image	ARIS polar frames	`aris_polar/`
`.jpg`	Image	GoPro camera frames	`gopro/`
`.csv`	Tabular	Frame metadata	`aris_frame_meta.csv`
`.csv`	Tabular	Gantry positions	`gantry.csv`
`.csv`	Tabular	AR3 poses	`ar3.csv`
`.yaml`	Config	File metadata	`aris_file_meta.yaml`
`.yaml`	Config	Calibration	`calibration/transforms.yaml`
`.txt`	Text	Recording notes	`notes.txt`

Working with the Data

Python Example

import os
import cv2
import yaml
import pandas as pd

# Load recording data
recording_dir = "recordings/100lbs/recording_001"

# Load ARIS frames
aris_frame = cv2.imread(
    os.path.join(recording_dir, "aris_polar/0042.png"), 
    cv2.IMREAD_GRAYSCALE
)

# Load GoPro frame
gopro_frame = cv2.imread(
    os.path.join(recording_dir, "gopro/0042.jpg")
)

# Load metadata
frame_meta = pd.read_csv(
    os.path.join(recording_dir, "aris_frame_meta.csv"),
    index_col="FrameIndex"
)
gantry_data = pd.read_csv(
    os.path.join(recording_dir, "gantry.csv"),
    index_col="aris_frame_idx"
)
ar3_pose = pd.read_csv(
    os.path.join(recording_dir, "ar3.csv"),
    index_col="aris_frame_idx"
)

# Load file metadata
with open(os.path.join(recording_dir, "aris_file_meta.yaml")) as f:
    file_meta = yaml.safe_load(f)

# Access specific frame data
frame_42_meta = frame_meta.loc[42]
print(f"Sound speed: {frame_42_meta['SoundSpeed']} m/s")
print(f"Gantry position: {gantry_data.loc[42].values}")

Viewing Tool

The dataset includes a viewer script for synchronized playback:

python scripts/view_recording.py recordings/100lbs/<recording-folder>

Use arrow keys to navigate frames. See view_recording.py for implementation.

Next Steps

Recording Structure

Learn about the organization of recording folders

Target Details

Explore calibration and target specifications

​Image Formats

​PGM - ARIS Raw Frames

​PNG - ARIS Polar Frames

​JPG - GoPro Camera Frames

​Tabular Data Formats

​CSV - Metadata and Poses

​aris_frame_meta.csv

​gantry.csv

​ar3.csv

​YAML - Calibration and Metadata

​aris_file_meta.yaml

​transforms.yaml

​Text Formats

​notes.txt

​Binary Formats

​ARIS Frame Data Structure

​Format Summary Table

​Working with the Data

​Python Example

​Viewing Tool

​Next Steps

Recording Structure

Target Details

Image Formats

PGM - ARIS Raw Frames

PNG - ARIS Polar Frames

JPG - GoPro Camera Frames

Tabular Data Formats

CSV - Metadata and Poses

aris_frame_meta.csv

gantry.csv

ar3.csv

YAML - Calibration and Metadata

aris_file_meta.yaml

transforms.yaml

Text Formats

notes.txt

Binary Formats

ARIS Frame Data Structure

Format Summary Table

Working with the Data

Python Example

Viewing Tool

Next Steps