Clément Chatelain

Tiramisu's tiered transformer architecture revolutionizes newspaper image understanding by explicitly modeling document hierarchy, outperforming traditional methods in reconstructing complex layouts.

Mamba-based OCR can process long paragraphs 1.4 to 4.5 times faster than Transformers, but struggles with handwriting due to data limitations.

Forget quadratic complexity: Mamba-based OCR rivals Transformer accuracy while halving inference time and scaling memory more efficiently, finally making paragraph-level historical text transcription practical.