PiTL: Cross-modal Retrieval with Weakly-supervised Vision-language Pre-training via Prompting
This work addresses the need for less supervised data in vision-language pre-training, which is incremental by building on weakly-supervised approaches.
The paper tackles the problem of reducing supervision in vision-language pre-training by proposing PiTL, which uses large language models to generate image descriptions from category labels, creating a new dataset of 9M images and 1M descriptions. The result shows that models pre-trained with PiTL-generated pairs outperform other weakly-supervised methods on cross-modal retrieval tasks with less supervision.
Vision-language (VL) Pre-training (VLP) has shown to well generalize VL models over a wide range of VL downstream tasks, especially for cross-modal retrieval. However, it hinges on a huge amount of image-text pairs, which requires tedious and costly curation. On the contrary, weakly-supervised VLP (W-VLP) explores means with object tags generated by a pre-trained object detector (OD) from images. Yet, they still require paired information, i.e. images and object-level annotations, as supervision to train an OD. To further reduce the amount of supervision, we propose Prompts-in-The-Loop (PiTL) that prompts knowledge from large language models (LLMs) to describe images. Concretely, given a category label of an image, e.g. refinery, the knowledge, e.g. a refinery could be seen with large storage tanks, pipework, and ..., extracted by LLMs is used as the language counterpart. The knowledge supplements, e.g. the common relations among entities most likely appearing in a scene. We create IN14K, a new VL dataset of 9M images and 1M descriptions of 14K categories from ImageNet21K with PiTL. Empirically, the VL models pre-trained with PiTL-generated pairs are strongly favored over other W-VLP works on image-to-text (I2T) and text-to-image (T2I) retrieval tasks, with less supervision. The results reveal the effectiveness of PiTL-generated pairs for VLP.