textual alignment

New framework improves video captioning with claim-level rewards

Researchers have introduced Claim-Level Rubric Rewards (CuRe), a new framework for reinforcement learning in dense video captioning. This approach aims to overcome limitations of existing reward designs, such as holisti…