arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2007.12130cs.CVcs.LGcs.SDeess.AS
收藏

Sound2Sight:从声音和上下文生成视觉动态

Sound2Sight: Generating Visual Dynamics from Sound and Context

Mitsubishi Electric Research Laboratories(三菱电机研究实验室);University of Illinois at Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校)

Anoop Cherian, Moitreya Chatterjee, Narendra Ahuja

摘要:

跨模态的学习关联对于鲁棒的多模态推理至关重要,特别是在推理过程中可能缺少模态时。在本文中,我们研究这个问题的背景下,音频条件的视觉合成-一个任务,这是很重要的,例如,在遮挡推理。具体来说,我们的目标是生成未来的视频帧和它们的运动动态条件的音频和一些过去的帧。为了解决这个问题,我们提出了Sound 2Sight,这是一个深度变分框架,它被训练来学习基于音频和过去帧的联合嵌入的每帧随机先验。这种嵌入是通过一个多头注意力为基础的视听Transformer编码器学习。然后对所学习的先验进行采样,以进一步调节视频预测模块以生成未来帧。随机先验允许模型对与所提供的音频和过去上下文一致的多个可能的未来进行采样。此外,为了提高所生成的帧的质量和一致性,我们提出了一个多模式的视频,区分合成和一个真正的视听剪辑。我们在两个新的数据集上实证评估了我们的方法,即(i)带有惊喜障碍的多模态随机移动MNIST,(ii)Youtube绘画;以及现有的Audio-Set Drums数据集。我们广泛的实验表明,Sound 2Sight在生成的视频质量方面明显优于最先进的技术,同时还能生成多样化的视频内容。

英文摘要:

Learning associations across modalities is critical for robust multimodal reasoning, especially when a modality may be missing during inference. In this paper, we study this problem in the context of audio-conditioned visual synthesis -- a task that is important, for example, in occlusion reasoning. Specifically, our goal is to generate future video frames and their motion dynamics conditioned on audio and a few past frames. To tackle this problem, we present Sound2Sight, a deep variational framework, that is trained to learn a per frame stochastic prior conditioned on a joint embedding of audio and past frames. This embedding is learned via a multi-head attention-based audio-visual transformer encoder. The learned prior is then sampled to further condition a video forecasting module to generate future frames. The stochastic prior allows the model to sample multiple plausible futures that are consistent with the provided audio and the past context. Moreover, to improve the quality and coherence of the generated frames, we propose a multimodal discriminator that differentiates between a synthesized and a real audio-visual clip. We empirically evaluate our approach, vis-á-vis closely-related prior methods, on two new datasets viz. (i) Multimodal Stochastic Moving MNIST with a Surprise Obstacle, (ii) Youtube Paintings; as well as on the existing Audio-Set Drums dataset. Our extensive experiments demonstrate that Sound2Sight significantly outperforms the state of the art in the generated video quality, while also producing diverse video content.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑