VLRM: Vision-Language Models act as Reward Models for Image Captioning
This work addresses the challenge of improving image captioning quality for applications in accessibility and content generation, though it is incremental as it builds on existing models like BLIP2.
The paper tackles the problem of generating more comprehensive image captions by using vision-language models as reward models in an unsupervised reinforcement learning framework, achieving a 0.90 R@1 CLIP Recall score on the MS-COCO Carpathy Test Split.
In this work, we present an unsupervised method for enhancing an image captioning model (in our case, BLIP2) using reinforcement learning and vision-language models like CLIP and BLIP2-ITM as reward models. The RL-tuned model is able to generate longer and more comprehensive descriptions. Our model reaches impressive 0.90 R@1 CLIP Recall score on MS-COCO Carpathy Test Split. Weights are available at https://huggingface.co/sashakunitsyn/vlrm-blip2-opt-2.7b.