Text-Only Training for Image Captioning using Noise-Injected CLIP

Autor:	Nukrai, David, Mokady, Ron, Globerson, Amir
Rok vydání:	2022
Předmět:	Computer Science - Computer Vision and Pattern Recognition Computer Science - Artificial Intelligence Computer Science - Machine Learning
Zdroj:	EMNLP 2022
Druh dokumentu:	Working Paper
DOI:	10.48448/n7sq-p557
Popis:	We consider the task of image-captioning using only the CLIP model and additional text data at training time, and no additional captioned images. Our approach relies on the fact that CLIP is trained to make visual and textual embeddings similar. Therefore, we only need to learn how to translate CLIP textual embeddings back into text, and we can learn how to do this by learning a decoder for the frozen CLIP text encoder using only text. We argue that this intuition is "almost correct" because of a gap between the embedding spaces, and propose to rectify this via noise injection during training. We demonstrate the effectiveness of our approach by showing SOTA zero-shot image captioning across four benchmarks, including style transfer. Code, data, and models are available on GitHub. Comment: Will be presented at EMNLP 2022. GitHub: https://github.com/DavidHuji/CapDec
Databáze:	arXiv
Externí odkaz:	http://arxiv.org/abs/2211.00575 Zobrazit plný text záznamu View this record from Arxiv