minseojung/PRISM-Dataset

PRISM Dataset

Dataset Description

The PRISM Dataset is a multi-modal automotive dataset collected using polarimetric cameras, LiDAR, and RGB cameras.

Dataset Structure

This repository contains raw sensor data organized by scenes and sequences:

dataset_name/
├── sequence_001/
│   ├── polar/           # Polarimetric images (4 angles)
│   ├── rgb/             # RGB camera images
│   └── lidar_single_scan/  # LiDAR point clouds
├── sequence_002/
│   └── ...

Included Modalities:

• polar/: Polarimetric camera images at 4 angles (0°, 45°, 90°, 135°)
  • Subfolders: 0d/, 45d/, 90d/, 135d/
  • Each contains grayscale images with Bayer pattern
  • Use for computing AoLP (Angle of Linear Polarization) and DoLP (Degree of Linear Polarization)
• rgb/: RGB camera images (PNG format)
• lidar_single_scan/: Single-frame LiDAR point clouds (.bin format)
  • Binary files with [x, y, z, intensity] per point

Note: Processed data (depth maps, elevation maps, vehicle state) are not included in this upload. Only raw sensor data is provided.

Data Collection Environments

The dataset includes 20 scenes across various driving scenarios:

Urban Roads:

• Hanyang University (HY): Clear weather, Snow
• Konkuk University: Clear weather, Snow, Multiple scenes
• Hanjayeon: 7 different scenes

Highway:

• Gapyeong: 2 scenes

Test Track:

• K-City: Clear weather (3 scenes + road scene), Rainy weather (2 scenes)

Sensor Setup

• Polarimetric Camera: 4-angle polarization (0°, 45°, 90°, 135°)
• RGB Camera: Standard color camera
• LiDAR: 3D point cloud sensor
• GPS/IMU: Vehicle positioning and motion data

Dataset Statistics

• Total Scenes: 20
• Weather Conditions: Clear, Snow, Rainy
• Locations: University campuses, Highway, Test track

Usage

Loading the Dataset

from datasets import load_dataset
from pathlib import Path

# Load dataset
dataset = load_dataset("minseojung/PRISM-Dataset")

# Or access files directly
# Example: Access polar images from a specific sequence
polar_0d_path = "0106_HY_dataset/sequence_001/polar/0d/"
rgb_path = "0106_HY_dataset/sequence_001/rgb/"
lidar_path = "0106_HY_dataset/sequence_001/lidar_single_scan/"

Computing Polarization Parameters

import cv2
import numpy as np

def compute_stokes(img_0d, img_45d, img_90d, img_135d):
    """Compute Stokes parameters from 4-angle polarization."""
    s0 = 0.5 * (img_0d + img_90d + img_45d + img_135d)
    s1 = img_0d - img_90d
    s2 = img_45d - img_135d
    return s0, s1, s2

def compute_aolp_dolp(img_0d, img_45d, img_90d, img_135d):
    """Compute AoLP and DoLP from 4-angle images."""
    s0, s1, s2 = compute_stokes(img_0d, img_45d, img_90d, img_135d)
    dolp = np.sqrt(s1**2 + s2**2) / (s0 + 1e-7)  # Degree of Linear Polarization
    aolp = 0.5 * np.arctan2(s2, s1)  # Angle of Linear Polarization
    return aolp, dolp

# Load 4-angle images
img_0d = cv2.imread("polar/0d/frame_000.png", cv2.IMREAD_GRAYSCALE).astype(np.float32) / 255.0
img_45d = cv2.imread("polar/45d/frame_000.png", cv2.IMREAD_GRAYSCALE).astype(np.float32) / 255.0
img_90d = cv2.imread("polar/90d/frame_000.png", cv2.IMREAD_GRAYSCALE).astype(np.float32) / 255.0
img_135d = cv2.imread("polar/135d/frame_000.png", cv2.IMREAD_GRAYSCALE).astype(np.float32) / 255.0

# Compute polarization parameters
aolp, dolp = compute_aolp_dolp(img_0d, img_45d, img_90d, img_135d)

Reading LiDAR Data

import numpy as np

def load_lidar_bin(bin_path):
    """Load LiDAR point cloud from .bin file."""
    # Each point: [x, y, z, intensity] as float32
    points = np.fromfile(bin_path, dtype=np.float32).reshape(-1, 4)
    return points

# Example
points = load_lidar_bin("lidar_single_scan/frame_000.bin")
xyz = points[:, :3]  # 3D coordinates
intensity = points[:, 3]  # Intensity values

Applications

This dataset can be used for:

• Depth Estimation: Using polarimetric images and LiDAR data
• 3D Reconstruction: Multi-modal sensor fusion
• Autonomous Driving: Perception in various weather conditions
• Polarization-based Vision: Research on polarimetric imaging

File Formats

• Polar Images: PNG format (grayscale with Bayer pattern)
  • Resolution: 1224 x 1024
  • 4 angle folders: 0d, 45d, 90d, 135d
• RGB Images: PNG format (color)
  • Resolution: 1224 x 1024
• LiDAR: Binary format (.bin)
  • Structure: [x, y, z, intensity] per point (float32)
  • Coordinate system: Vehicle frame

Citation

If you use this dataset in your research, please cite:

@dataset{prism2025,
  title={PRISM Dataset: Multi-modal Automotive Dataset with Polarimetric Camera},
  author={Jung, Minseo and others},
  year={2025},
  publisher={Hugging Face}
}

License

[Specify your license here - e.g., CC BY 4.0, MIT, etc.]

Contact

For questions or issues, please open an issue on the repository.

---

🤖 Dataset prepared and uploaded using Python scripts.