Add WebP Image Dataset Support for HDF5 Storage Format

[NetZissou]

WebP Images in HDF5 Dataset Class

Overview:

Currently, the datasets module supports:

• Parquet datasets
• Image folder datasets

Add a new dataset class to handle WebP images stored in HDF5 format, which would provide:

• Efficient storage and retrieval of WebP compressed images
• Better I/O performance for large image collections
• Seamless integration with existing batch inference pipeline

Use Cases

• Processing large collections of WebP images stored in HDF5 archives
• Efficient batch inference on compressed image datasets
• Scientific computing workflows requiring high-performance image I/O

Requirements& Todos

• Create WebPHDF5Dataset class in the datasets module
• Support for reading WebP images from HDF5 groups/datasets
• Integration with existing DataLoader infrastructure
• Proper error handling for corrupted or missing images
• Memory-efficient streaming for large datasets

Dependencies:

• h5py for HDF5 file operations
• pillow or opencv-python for WebP decoding
• Existing dataset infrastructure

Documentation:

• API documentation
• Usage examples
• Performance comparison guide