Hub v2.3.4 introduces support for popular image formats in hub.ingest and hub.ingest_kaggle, enabling direct ingestion of datasets from Kaggle. A new ds.summary() function provides an overview of dataset layouts. PyTorch Dataloaders can now return data as bytes instead of tensors, allowing for custom decompression and removal processes. Less intrusive locking is implemented when performing operations on different version control branches. Community contributions include new datasets from Uday Uppal and Manas Gupta, str return improvements by Suhaas Neel, additional image format support in hub.auto by Bikram Maharjan, and a Chinese ReadMe translation by Jinyi Chen. Gradient Health has published a guide to Open-source Medical Imagery Datasets.