Stardust SDK

Stardust SDK helps data and machine learning teams orchestrate the lifecycle of high quality sensor datasets: exporting annotations, converting formats, validating metrics, and closing the loop on model performance.

Overview

Stardust-SDK is a Python toolkit created by Stardust.ai for preparing training data. It connects raw annotations from Rosetta and MorningStar to common computer vision formats, analytics, and visualization utilities so that models can be trained and debugged faster.

Features

• Export labeled data from Rosetta or MorningStar in the Stardust data format and transform it into COCO, KITTI, or PandaSet with minimal code.
• Visualize predictions for key perception tasks such as object detection, tracking, and lane detection.
• Evaluate model quality with built-in metrics including precision, recall, and F1, or plug in custom callbacks.
• Validate camera and lidar calibration matrices before they reach production pipelines.

Installation

Option 1: Install from PyPI (Recommended)

pip install stardust-sdk

Option 2: Install from GitHub

pip install git+https://github.com/stardustai/Stardust-SDK.git

Option 3: Development Installation

1. Create and activate a Python 3.9 (or newer) environment.

   conda create -n stardust_sdk python=3.9 -y
   conda activate stardust_sdk
   

2. Clone this repository.

   git clone https://github.com/stardustai/Stardust-SDK.git
   cd Stardust-SDK
   

3. Install dependencies and the package in editable mode.

   pip install -r requirements.txt
   pip install -e .
   

Quick Start

Export Stardust JSON

Use read_rosetta to load labeled frames from Rosetta and stream them to a JSON file.

import json
from stardust.utils.convert import read_rosetta

gen_data = read_rosetta(
    project_id=1694,
    input_path="/path/to/export",
    pool_lst=[48904],  # optional
    export_type="json"  # optional
)

with open("your_file.json", "w", encoding="utf-8") as sf:
    for json_data in gen_data:
        json.dump(json_data, sf, ensure_ascii=False)
        break

Convert to KITTI

Transform Stardust JSON into KITTI files with the high level conversion helper.

import os
from stardust.utils.convert import read_rosetta
from stardust.convertion.to_kitti.to_kitti import kitti_export


gen_json_data = read_rosetta(
    project_id=1507,
    input_path="/path/to/export"
)

kitti_export(
    gen_json_data,
    os.path.join("/path/to/export", "kitti")
)  # outputs into input_path/kitti

Compute Metrics

Compare model predictions against human labels and review the results.

from pprint import pprint
from stardust.utils.convert import read_rosetta
from stardust.metric.object_detection import compute_metric

project_id = 1507
json_datas = read_rosetta(project_id=project_id, input_path="/path/to/export")
metric = compute_metric(json_datas, 0.5, "IoU", "results")
pprint(metric)

Project 3D Points

Confirm calibration by projecting 3D point clouds onto camera images.

import numpy as np
import open3d as o3d
from stardust.geometry.ops3d import pointcloud2img
from stardust.utils.convert import read_rosetta


pcd_path = "/path/to/point_cloud.pcd"
frames_jsons = read_rosetta(
    project_id=1507,
    input_path="/path/to/export",
    export_type="json"
)

pcd = o3d.io.read_point_cloud(pcd_path)
pcd_pts = np.asarray(pcd.points)

for frame in frames_jsons:
    frame_info = frame["media"]
    img_pixel = pointcloud2img(frame_info, cam_idx=1, points_3d=pcd_pts)
    # img_pixel now contains the projection of the point cloud onto image 1

Data Model

Each exported frame contains synchronized metadata and annotations. The top level structure is:

{
  "media": "Media metadata for all sensor files",
  "task_info": "TaskInfo describing the labeling job context",
  "annotation": "Human annotation results",
  "prediction": "Model-generated pre-annotations"
}

Key components are documented in the API reference:

• Media: docs/statics/frame.html#stardust.components.frame.Media
• Annotation: docs/statics/frame.html#stardust.components.frame.Annotation
• TaskInfo: docs/statics/frame.html#stardust.components.frame.TaskInfo

Refer to the documentation site for a complete schema and additional examples.

Documentation

Latest API and workflow guides: docs.

Contributing

We welcome pull requests and feature proposals. Please review the contributing guidelines before submitting code.

Roadmap

• Streamlined data warehouse migration support.
• Hard example discovery tooling for model debugging.
• Active learning workflow integration.

Community

Have questions or feedback? Join the discussion on Discord: https://discord.gg/gaaENMpK