The New Data and New Challenges in Multimedia Research

By Bart Thomee, David A. Shamma, Gerald Friedland, Benjamin Elizalde, Karl S. Ni, Douglas N. Poland, Damian Borth, Li-Jia Li. 2015

332 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Multimodal Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Multimodal Learning

Citations in the corpus

Revisiting Unreasonable Effectiveness of Data in Deep Learning Era (2017, methodology): “While [44] looked at geo-localization, [23] utilized the YFCC-100M dataset [40] for representation learning.”
Learning Visual Features from Large Weakly Supervised Data (2015, methodology): “We train our models on the publicly available Flickr 100M data set [47].”
Unsupervised Visual Representation Learning by Context Prediction (2015, methodology): “To understand the effect of various dataset biases [52], we also performed a preliminary experiment pre-training on a randomly-selected 2M subset of the Yahoo/Flickr 100million Dataset [51], which was collected entirely automatically.”
Scaling Laws for Autoregressive Generative Modeling (2020, methodology): “We show the same phenomena for language in ﬁgure 15, though we can only estimate 10 L ( D ) for these models.”
Deep Clustering for Unsupervised Learning of Visual Features (2018, methodology): “Table 3 shows the diﬀerence in performance on Pascal VOC of DeepClus-ter pre-trained on YFCC100M compared to ImageNet.”
Multimodal Compact Bilinear Pooling for Visual Question Answering and Visual Grounding (2016, methodology): “2016) uses 108,249 images from the intersection of YFCC100M (Thomee et al., 2015) and MSCOCO.”
Billion-scale semi-supervised learning for image classification (2019, background): “• YFCC-100M [38] is a publicly available dataset of about 90 million images from Flickr website with associated tags.”
Learning Sight from Sound: Ambient Sound Provides Supervision for Visual Learning (2017, neutral): “However, this is potentially suboptimal because in natural scenes it is difﬁcult to predict the precise timing of a sound from We show two frames from the YFCC100m dataset (Thomee et al 2015).”
DenseCap: Fully Convolutional Localization Networks for Dense Captioning (2015, methodology): “Images were taken from the intersection of MS COCO and YFCC100M [46], and annotations were collected on Amazon Mechanical Turk by asking workers to draw a bounding box on the image and describe its content in text.”
YouTube-8M: A Large-Scale Video Classification Benchmark (2016, background): “Currently, the largest available video benchmarks are the Sports-1M [19], with 487 sports related activities and 1 M videos, the YFCC-100M [34], with 800 K videos and raw metadata (titles, descriptions, tags) for some of them, the FCVID [17] dataset of 91 , 223 videos manually annotated with 239…”
Localizing Moments in Video with Natural Language (2017, methodology): “We randomly select over 14,000 videos from YFCC100M [44] which contains over 100,000 Flickr videos with a Creative Commons License.”
Seeing through the Human Reporting Bias: Visual Classifiers from Noisy Human-Centric Labels (2015, background): “Apart from just numerical improvements, our results are interpretable and consistent with research in psychology showing that humans tend not to mention typical attributes [31, 53] unless required for unique identification [44] or distinguishability [39, 48].”