Spectro-polarimetric Dataset

We provide a spectro-polarimetric dataset. This dataset consists of full-Stokes images for both hyperspectral and trichromatic scenes. Hyperspectral dataset has 311 scenes and trichromatic dataset has 2022 scenes.

For more details, see our paper on Spectral and Polarization Vision: Spectro-polarimetric Real-world Dataset.

⚠️ Notice: Known Issues in Hyperspectral Dataset

1. Measurement Errors Specific wavelengths and polarization components were not measured correctly. IDs: 089, 112 2. Polarization Misalignment Misalignment exists between x-polarization and y-polarization channels. IDs: 001, 029, 032, 047, 052, 059, 060, 061, 062, 201, 205, 269, 294, 307, 308 3. Misalignment due to Scene Motion Slight misalignment occurred due to object movement within the scene during capture. IDs: 223, 229, 285, 292, 293, 295, 306, 309, 310

📦 Contents

• File Hierarchy
• Trichromatic Data Overview
• Hyperspectral Data Overview
• Labeling Information
• Citation
• Contact

💾 File Hierarchy

• 📂 trichromatic/: Trichromatic polarimetric dataset

  • 📂 original/: Data processed from captured raw files
    • 📄 0000.npy ~
  • 📂 denoised/: Data processed from denoised raw files
    • 📄 0000.npy ~
  • 📂 mask/: Masks for the scenes
    • 📄 0000.png ~
  • 📄 labeling_trichromatic.csv: Labeling each scene

• 📂 hyperspectral/: Hyperspectral polarimetric dataset

  • 📂 original/: Data processed from captured raw files
    • 📄 0000.npy ~
  • 📂 denoised/: Data processed from denoised raw files
    • 📄 0000.npy ~
  • 📂 mask/: Mask fors the scenes
    • 📄 0000.png ~
  • 📄 labeling_hyperspectral.csv: Labeling each scene

• 📄 README.md

📷 Trichromatic Data Overview

• original & denoised:

  • Format: Stokes numpy files (1900, 2100, 4, 3)
  • Dimensions:
    • R: Spatial dimension
    • C: Spatial dimension
    • s: Stokes axis (s0, s1, s2, s3)
    • c: Spectral axis (R G B)

• mask:

  • Binary mask images (1900, 2100)
  • Dimensions:
    • R: Spatial dimension
    • C: Spatial dimension

📷 Hyperspectral Data Overview

• original & denoised:

  • Format: Stokes numpy files (512, 612, 4, 21)
  • Dimensions:
    • R: Spatial dimension
    • C: Spatial dimension
    • s: Stokes axis (s0, s1, s2, s3)
    • c: Spectral axis (450nm ~ 650nm at 10nm intervals)

• mask:

  • Binary mask images (512, 612)
  • Dimensions:
    • R: Spatial dimension
    • C: Spatial dimension

🗂️ Labeling Information

• SceneNum: Indices of scenes (0 ~ 310 for hyperspectral, 0 ~ 2021 for trichromatic)
• Content: Type of content (scene, object)
• Timezone: Scene environment (indoor, day, night)
• Illumination: Illumination condition (white, yellow, sunny, cloudy)

📜 Citation

@InProceedings{Jeon_2024_CVPR,
    author    = {Jeon, Yujin and Choi, Eunsue and Kim, Youngchan and Moon, Yunseong and Omer, Khalid and Heide, Felix and Baek, Seung-Hwan},
    title     = {Spectral and Polarization Vision: Spectro-polarimetric Real-world Dataset},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024},
    pages     = {22098-22108}
}

📫 Contact

If you have any questions, please contact

• Yujin Jeon: [email protected]
• Eunsue Choi: [email protected]
• Seung-hwan Baek: [email protected]