Factorized vector quantization matches continuous visual tokens on understanding
The discrete model beats UniAR and Emu3-Chat by 22.8% and 59.2% on a nine-metric average.
Stories per day · 30 days
2 RSS
Relatedimage-generation1knowledge-transfer1ocr1question-answering1representation-learning1vector-quantization1visual-tokens1
2 stories, newest first · one card per story
Stories tagged #multimodal-models
Paper tests whether image generation expands model knowledge
Across seven knowledge categories and three unified models, visual supervision transferred unevenly into text-only question answering.