arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2209.01540cs.CV
收藏

采用掩码视觉建模的端到端视频-语言 Transformer 实证研究

An Empirical Study of End-to-End Video-Language Transformers with Masked Visual Modeling

UC Santa Barbara(加州大学圣巴巴拉分校);Microsoft(微软)

Tsu-Jui Fu, Linjie Li, Zhe Gan, Kevin Lin, William Yang Wang, Lijuan Wang, Zicheng Liu

AI总结:

本文基于端到端 VIOLET 系统研究八种 MVM 重建目标,揭示有效训练因素并提出 VIOLETv2,在 13 个视频-语言基准上显著提升性能。

摘要:

掩码视觉建模(MVM)近来已被证明对视觉预训练有效。尽管在视频-语言(VidL)预训练中已经探索了针对视频输入的类似重建目标(例如掩码帧建模),但以往研究未能找到一种真正有效、能够显著提升下游性能的 MVM 策略。在本工作中,我们系统考察了 MVM 在 VidL 学习场景中的潜力。具体而言,我们的研究基于完全端到端的 VIdeO-LanguagE Transformer(VIOLET),其中来自 MVM 训练的监督可以反向传播到视频像素空间。研究共探索了八种不同的 MVM 重建目标,从低层像素值和方向梯度,到高层深度图、光流、离散视觉 token 以及潜在视觉特征。我们开展了全面实验,并对促成有效 MVM 训练的因素给出洞见,由此得到增强模型 VIOLETv2。实验表明,使用 MVM 目标预训练的 VIOLETv2 在 13 个 VidL 基准上取得显著改进,任务涵盖视频问答、视频字幕生成以及文本到视频检索。

英文摘要:

Masked visual modeling (MVM) has been recently proven effective for visual pre-training. While similar reconstructive objectives on video inputs (e.g., masked frame modeling) have been explored in video-language (VidL) pre-training, previous studies fail to find a truly effective MVM strategy that can largely benefit the downstream performance. In this work, we systematically examine the potential of MVM in the context of VidL learning. Specifically, we base our study on a fully end-to-end VIdeO-LanguagE Transformer (VIOLET), where the supervision from MVM training can be backpropagated to the video pixel space. In total, eight different reconstructive targets of MVM are explored, from low-level pixel values and oriented gradients to high-level depth maps, optical flow, discrete visual tokens, and latent visual features. We conduct comprehensive experiments and provide insights into the factors leading to effective MVM training, resulting in an enhanced model VIOLETv2. Empirically, we show VIOLETv2 pre-trained with MVM objective achieves notable improvements on 13 VidL benchmarks, ranging from video question answering, video captioning, to text-to-video retrieval.

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。

↑