Masking Modalities for Cross-modal Video Retrieval
This work addresses the problem of improving video retrieval for researchers and practitioners by proposing a novel pre-training method, though it is incremental as it builds on existing weak supervision strategies.
The paper tackles the problem of pre-training video encoders by addressing the limitation of using only speech as weak supervision, which fails to exploit other modalities like appearance and audio. The result is a modality masking approach that achieves superior performance for cross-modal video retrieval on datasets such as How2R, YouCook2, and Condensed Movies.
Pre-training on large scale unlabelled datasets has shown impressive performance improvements in the fields of computer vision and natural language processing. Given the advent of large-scale instructional video datasets, a common strategy for pre-training video encoders is to use the accompanying speech as weak supervision. However, as speech is used to supervise the pre-training, it is never seen by the video encoder, which does not learn to process that modality. We address this drawback of current pre-training methods, which fail to exploit the rich cues in spoken language. Our proposal is to pre-train a video encoder using all the available video modalities as supervision, namely, appearance, sound, and transcribed speech. We mask an entire modality in the input and predict it using the other two modalities. This encourages each modality to collaborate with the others, and our video encoder learns to process appearance and audio as well as speech. We show the superior performance of our "modality masking" pre-training approach for video retrieval on the How2R, YouCook2 and Condensed Movies datasets.