README.md

metadata

language:
  - en
license: cc-by-4.0
task_categories:
  - image-to-image
pretty_name: SCaN-TIR

SCaN-TIR: Stereo Clean-and-Noisy Paired Thermal Infrared Dataset

TIDY: Thermal Infrared Image Denoising via Wavelet Domain Entropy and Directional Stripe Index

Code | Dataset | Video

SCaN-TIR is a real clean–noisy paired thermal infrared dataset for supervised TIR image denoising. It contains over 32.5K hardware-synchronized and geometrically aligned image pairs captured with two adjacent FLIR A65 thermal cameras, where one camera stream was intentionally degraded by disabling non-uniformity correction (NUC). The dataset includes both indoor and outdoor sequences at 640×512 resolution, providing real sensor noise rather than synthetic noise approximations.

Highlights

• First real stereo clean–noisy paired and aligned TIR dataset
• Over 32.5K synchronized thermal image pairs
• Captured using two adjacent FLIR A65 infrared cameras
• One stream collected with NUC disabled to obtain real noisy thermal images
• Geometrically aligned through stereo rectification using calibrated extrinsics
• Includes both indoor and outdoor sequences
• Provides both Fieldscale-mapped 8-bit images and original 14-bit images
• Designed for supervised real-noise TIR denoising and downstream robotics perception

Usage

You can download the dataset using the Hugging Face CLI:

pip install -U "huggingface_hub[cli]"

huggingface-cli download williamrhee/SCaN-TIR \
    --repo-type dataset \
    --local-dir SCaN-TIR

Dataset Structure

The dataset is organized into scene folders containing synchronized clean and noisy thermal infrared images.

SCaN-TIR
  ├── {$SEQUENCE_NAME}
  |   ├── left_clean
  |   |   ├── frame_00000.png
  |   |   └── ...
  |   ├── right_noisy
  |   |   ├── frame_00000.png
  |   |   └── ...
  |   ├── thermal_14bit_left_image_raw
  |   |   ├── frame_00000.png
  |   |   └── ...
  |   └── thermal_14bit_right_image_raw
  |       ├── frame_00000.png
  |       └── ...
  ├── ...
...

Citation

If you use SCaN-TIR in your research, please cite:

@inproceedings{rhee2026tidy,
  title={TIDY: Thermal Infrared Image Denoising via Wavelet Domain Entropy and Directional Stripe Index},
  author={Rhee, Tai Hyoung and Lee, Dong-guw and Kim, Ayoung},
  booktitle={IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year={2026},
  organization={IEEE}
}