VideoOFA: Two-Stage Pre-Training for Video-to-Text Generation
This work addresses the problem of generating accurate text from videos for applications like captioning and question answering, representing an incremental advance in pre-training methods for video-language models.
The authors tackled video-to-text generation by proposing a two-stage pre-training framework that first learns vision-language concepts from image-text data and then adapts to video data, achieving new state-of-the-art performance with an average 9.7-point CIDEr score improvement on video captioning benchmarks and outperforming models on video question answering.
We propose a new two-stage pre-training framework for video-to-text generation tasks such as video captioning and video question answering: A generative encoder-decoder model is first jointly pre-trained on massive image-text data to learn fundamental vision-language concepts, and then adapted to video data in an intermediate video-text pre-training stage to learn video-specific skills such as spatio-temporal reasoning. As a result, our VideoOFA model achieves new state-of-the-art performance on four Video Captioning benchmarks, beating prior art by an average of 9.7 points in CIDEr score. It also outperforms existing models on two open-ended Video Question Answering datasets, showcasing its generalization capability as a universal video-to-text model.