Text-Only Training for Image Captioning using Noise-Injected CLIP
This enables image captioning without costly labeled data, though it is incremental as it builds on CLIP.
The paper tackles image captioning without paired image-caption data by training a decoder for CLIP's text encoder using only text, enhanced with noise injection to address embedding gaps, achieving state-of-the-art zero-shot performance on four benchmarks.
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.