MLIC-Train-100K: Training Images for Learned Compression Models

A collection of 100,000 images for training learned image compression models, as referenced by the dataset title. The dataset was created by Whiteboat and last updated on May 13, 2025. The associated research abstract discusses capturing channel-wise, local spatial, and global spatial correlations in latent representations for entropy modeling.

Use Cases

• Training entropy models for learned image compression based on the described latent representation correlations.
• Benchmarking global context modules in high-resolution image coding scenarios mentioned in the abstract.
• Developing conditional entropy minimization techniques for image compression.
• Researching methods to efficiently capture channel-wise and spatial correlations in latent spaces.

Strengths

• Dataset title indicates a scale of 100,000 training images.
• Associated with a specific research paper abstract discussing entropy modeling for learned compression.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Whiteboat on Hugging Face

Freshness

Last updated 2025-05-13 11:21:53; freshness should be verified.