Support More Training Data Formats

It would be helpful to support a wider variety of data formats for training. Suggested improvement Allow users to provide training data in multiple commonly used formats, with automatic detection and parsing where possible. For example: Images with metadata/labels Other commonly used dataset formats Clear documentation for each supported format and its required structure would also make dataset preparation easier. The goal is to make the training workflow more flexible by allowing users to use their existing datasets without having to manually convert them into a single specific format.