ariakang/ADT-test

ADT Dataset

Dataset Description

This dataset contains Aria Digital Twin (ADT) sequences with various sensor data and annotations, including 2D/3D bounding boxes, trajectories, eye gaze data, and VRS recordings.

Quick Start

from adt_dataset_loader import ADTDatasetLoader

# Load entire dataset
loader = ADTDatasetLoader("ariakang/ADT-test")

# Load specific sequence
loader = ADTDatasetLoader("ariakang/ADT-test", sequence_name="Apartment_release_clean_seq131_M1292")

Installation

# Install required packages
pip install datasets pandas

Dataset Structure

Each sequence contains:

• VRS Files:
  • video.vrs
  • synthetic_video.vrs
  • segmentations.vrs
  • depth_images.vrs
• CSV Data:
  • 2D/3D bounding boxes
  • Aria device trajectories
  • Eye gaze data
  • Scene objects
• JSON Data:
  • Instance annotations
  • Metadata
• MPS Data:
  • Eye gaze processing
  • SLAM results

Flexible Loading Options

1. Load Entire Dataset

# Initialize loader with all sequences
loader = ADTDatasetLoader("ariakang/ADT-test")

# See available sequences and data types
available_files = loader.get_available_files()
print("Available files:", available_files)

# Load all data types
bbox_2d = loader.load_2d_bounding_boxes()
bbox_3d = loader.load_3d_bounding_boxes()
trajectory = loader.load_aria_trajectory()
eyegaze = loader.load_eyegaze()
metadata = loader.load_metadata()
slam_data = loader.load_mps_slam()

2. Load Specific Sequences

# Load a specific sequence
loader = ADTDatasetLoader(
    "ariakang/ADT-test",
    sequence_name="Apartment_release_clean_seq131_M1292"
)

# Load data from this sequence
bbox_2d = loader.load_2d_bounding_boxes()
trajectory = loader.load_aria_trajectory()

3. Load Selected Data Types

# Initialize loader for specific sequence
loader = ADTDatasetLoader("ariakang/ADT-test", "Apartment_release_clean_seq131_M1292")

# Load only 2D bounding boxes and VRS info
bbox_2d = loader.load_2d_bounding_boxes()
vrs_info = loader.get_vrs_files_info()

# Get paths to specific VRS files
video_vrs = [f for f in vrs_info if f['filename'] == 'video.vrs'][0]
print(f"Video VRS path: {video_vrs['path']}")

# Load only SLAM data
slam_data = loader.load_mps_slam()
closed_loop = slam_data['closed_loop']  # Get specific SLAM component

Available Data Types and Methods

Main Data Types

# Bounding Boxes and Trajectories
bbox_2d = loader.load_2d_bounding_boxes()
bbox_3d = loader.load_3d_bounding_boxes()
trajectory = loader.load_aria_trajectory()

# Eye Gaze and Scene Data
eyegaze = loader.load_eyegaze()
scene_objects = loader.load_scene_objects()

# Metadata and Instances
metadata = loader.load_metadata()
instances = loader.load_instances()

# MPS Data
eye_gaze_data = loader.load_mps_eye_gaze()  # Returns dict with 'general' and 'summary'
slam_data = loader.load_mps_slam()  # Returns dict with various SLAM components

VRS Files

# Get VRS file information
vrs_info = loader.get_vrs_files_info()

# Example: Access specific VRS file info
for vrs_file in vrs_info:
    print(f"File: {vrs_file['filename']}")
    print(f"Path: {vrs_file['path']}")
    print(f"Size: {vrs_file['size_bytes'] / 1024 / 1024:.2f} MB")

Custom Loading

# Load any file by name
data = loader.load_file_by_name("your_file_name.csv")

Data Format Examples

2D Bounding Boxes

bbox_2d = loader.load_2d_bounding_boxes()
print(bbox_2d.columns)
# Columns: ['object_uid', 'timestamp[ns]', 'x_min[pixel]', 'x_max[pixel]', 'y_min[pixel]', 'y_max[pixel]']

Aria Trajectory

trajectory = loader.load_aria_trajectory()
print(trajectory.columns)
# Columns: ['timestamp[ns]', 'x', 'y', 'z', 'qx', 'qy', 'qz', 'qw']

MPS SLAM Data

slam_data = loader.load_mps_slam()
# Components:
# - closed_loop: DataFrame with closed-loop trajectory
# - open_loop: DataFrame with open-loop trajectory
# - calibration: Calibration parameters

Error Handling

try:
    data = loader.load_file_by_name("non_existent_file.csv")
except ValueError as e:
    print(f"Error: {e}")

Notes

• All CSV files are loaded as pandas DataFrames
• JSON/JSONL files are loaded as Python dictionaries/lists
• VRS files are not loaded into memory, only their metadata and paths are provided
• Use get_available_files() to see all available data in your sequence

Repository Structure

VRS files are stored in sequence-specific folders: sequences/{sequence_name}/vrs_files/