YouTube-8M: A Large-Scale Video Classification Benchmark
Sami Abu-El-Haija Nisarg Kothari Joonseok Lee Paul Natsev George Toderici Balakrishnan Varadarajan Sudheendra Vijayanarasimhan Google Research
Abstract
Many recent advancements in Computer Vision are attributed to large datasets. Open-source software packages for Machine Learning and inexpensive commodity hardware have reduced the barrier of entry for exploring novel approaches at scale. It is possible to train models over millions of examples within a few days. Although large-scale datasets exist for image understanding, such as ImageNet, there are no comparable size video classification datasets.
中文速览
视频理解领域长期缺乏像ImageNet那样大规模、多样化的标注数据集，严重制约了深度学习方法的发展。为此，谷歌研究团队构建并发布了YouTube-8M——一个包含约800万个视频、覆盖500万小时、涵盖4800个视觉实体标签的超大规模多标签视频分类数据集（multi-label video classification dataset），标注来源于YouTube的自动话题标注系统，并经过人工审核以确保标签的视觉可识别性。为了降低使用门槛，他们用预训练的Inception网络以每秒一帧的速率提取视频帧特征，再经PCA降维和量化压缩，最终公开了近19亿帧的特征向量，让普通研究团队无需处理原始视频即可直接训练模型。实验表明，在YouTube-8M上预训练的模型迁移到ActivityNet等其他基准后，mAP从53.8%大幅提升至77.6%，验证了大规模预训练的强大泛化能力；这一数据集的发布有望成为推动视频理解和表示学习（representation learning）研究的重要基础设施。
原文 arXiv:1609.08675;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1609.08675v1