BriefGPT.xyz
Oct, 2022
面向持续视觉语言预训练的生成式负文本重播
Generative Negative Text Replay for Continual Vision-Language Pretraining
HTML
PDF
Shipeng Yan, Lanqing Hong, Hang Xu, Jianhua Han, Tinne Tuytelaars...
TL;DR
本研究针对连续多模态学习中的遗忘问题,通过伪文本回放和多模态知识蒸馏的方法,实现了基于图像和文本对的连续预训练,大幅提高了零样本图像分类和图像-文本检索任务的性能。
Abstract
vision-language pre-training
(
vlp
) has attracted increasing attention recently. With a large amount of image-text pairs,
vlp
models traine
→