End-to-end Generative Pretraining for Multimodal Video Captioning
Abstract
Recent video and language pretraining frameworks lack the ability to generate sentences. We present Multimodal Video Generative Pretraining (MV-GPT), a new pretraining framework for learning from unlabelled videos which can be effectively used for generative tasks such as multimodal video captioning. Unlike recent video-language pretraining frameworks, our framework trains both a multimodal video encoder and a sentence decoder jointly. To overcome the lack of captions in unlabelled videos, we leverage the future utterance as an additional text source and propose a bidirectional generation objective -- we generate future utterances given the present mulitmodal context, and also the present utterance given future observations. With this objective, we train an encoder-decoder model end-to-end to generate a caption from raw pixels and transcribed speech directly. Our model achieves state-of-the-art performance for multimodal video captioning on four standard benchmarks, as well as for other video understanding tasks such as VideoQA, video retrieval and action classification.
- Publication:
-
arXiv e-prints
- Pub Date:
- January 2022
- DOI:
- 10.48550/arXiv.2201.08264
- arXiv:
- arXiv:2201.08264
- Bibcode:
- 2022arXiv220108264H
- Keywords:
-
- Computer Science - Computer Vision and Pattern Recognition;
- Computer Science - Artificial Intelligence;
- Computer Science - Computation and Language;
- Computer Science - Human-Computer Interaction
- E-Print:
- Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR) 2022