metadata

license: mit
datasets:
  - nlphuji/flickr30k
language:
  - en

Dataset Card for Conditional Latent Coding (CLC)

Dataset Description

Repository: GitHub - ydchen0806/CLC
Paper: Conditional Latent Coding with Learnable Synthesized Reference for Deep Image Compression (AAAI25 Oral)
Authors: Siqi Wu†, Yinda Chen†, Dong Liu, Zhihai He*
Contact: [email protected]

Overview

This repository contains datasets and pre-trained models for the Conditional Latent Coding (CLC) framework, a state-of-the-art deep image compression method. The implementation is built on CompressAI and TCM.

Dataset Structure

Core Components

Reference Features (flicker_features.pkl):
- Precomputed feature dictionary using spatial pyramid pooling and k-means clustering
- Format: Pickle file containing clustered image features
Training Dataset (Flickr2K.hdf5):
- Contains 2,650 high-resolution images (256×256 patches)
- HDF5 structure:
```
/Flickr2K
    ├── image_0001
    ├── image_0002
    └── ...
```
Pre-trained Models:
- Multiple rate points (0.0025-0.05 bpp):
  - 0.0025checkpoint_best.pth.tar
  - 0.05checkpoint_best.pth.tar
- Compatibility: PyTorch 1.7+ with CUDA support

📜 Citation

If you use this model or find it useful, please cite:

@article{wu2025conditional,
  title={Conditional Latent Coding with Learnable Synthesized Reference for Deep Image Compression},
  author={Wu, Siqi and Chen, Yinda and Liu, Dong and He, Zhihai},
  journal={AAAI Conference on Artificial Intelligence},
  year={2025}
}

📧 Contact

For questions or collaborations, feel free to reach out:

GitHub: CLC Repository
Email: [email protected]