We are pleased to share that Fei Wu of the Computer Vision group has won second place in the “ICDAR 2026 Competition on Long-Term Handwriting Author Identification (Page Retrieval Track)”, where participants faced the difficult challenge of accurately identifying authorship across over 500,000 pages from 2,286 handwritten books. This makes exhaustive spatial search at every page computationally expensive. Fei’s solution aggregated multiple network embeddings obtained via random patch-level scanning into a single unified representation using an attention mechanism from Wu et al.’s previously developed AMD-HookNet++ semantic segmentation network. This holistic page embedding was ℓ₂-normalized and refined with a metric learning loss to learn a structured feature space for writer retrieval. He will receive the award at the 2026 International Conference on Document Analysis and Recognition, Vienna, Austria.
Pattern Recognition Lab Member wins Second Place in ICDAR Competition
