Distinctive Image Captioning via CLIP Guided Group Optimization
This work addresses the issue of generic captions in image captioning for applications requiring fine-grained image differentiation, representing an incremental improvement over existing methods.
The paper tackles the problem of generating distinctive image captions that can differentiate a target image from similar ones, introducing CLIP-based metrics to quantify distinctiveness and a training strategy that optimizes group embedding gaps, achieving new state-of-the-art results in distinctiveness.
Image captioning models are usually trained according to human annotated ground-truth captions, which could generate accurate but generic captions. In this paper, we focus on generating distinctive captions that can distinguish the target image from other similar images. To evaluate the distinctiveness of captions, we introduce a series of metrics that use large-scale vision-language pre-training model CLIP to quantify the distinctiveness. To further improve the distinctiveness of captioning models, we propose a simple and effective training strategy that trains the model by comparing target image with similar image group and optimizing the group embedding gap. Extensive experiments are conducted on various baseline models to demonstrate the wide applicability of our strategy and the consistency of metric results with human evaluation. By comparing the performance of our best model with existing state-of-the-art models, we claim that our model achieves new state-of-the-art towards distinctiveness objective.