[AAAI-25] ImagePiece: A Novel Content-Aware Re-Tokenization Technique for Semanticizing Tokens in Efficient ViT
Research paper
LG AI Research presented a paper at AAAI-25 introducing ImagePiece, a content-aware re-tokenization technique for Vision Transformers that re-tokenizes non-semantic visual tokens into semantic units, inspired by NLP's WordPiece, to improve efficiency and maintain expressiveness.
