Exploring the Limits of Weakly Supervised Pretraining

State-of-the-art visual perception models for a wide range of tasks rely on supervised pretraining. ImageNet classification is the de facto pretraining task for these models. Yet, ImageNet is now nearly ten years old and is by modern standards "small". Even so, relatively little is known about the behavior of pretraining with datasets that are multiple orders of magnitude larger. The reasons are obvious: such datasets are difficult to collect and annotate. In this paper, we present a unique study of transfer learning with large convolutional networks trained to predict hashtags on billions of social media images. Our experiments demonstrate that training for large-scale hashtag prediction leads to excellent results. We show improvements on several image classification and object detection tasks, and report the highest ImageNet-1k single-crop, top-1 accuracy to date: 85.4% (97.6% top-5). We also perform extensive experiments that provide novel empirical data on the relationship between large-scale pretraining and transfer learning performance.

PDF Abstract ECCV 2018 PDF ECCV 2018 Abstract

Datasets


Results from the Paper


Ranked #221 on Image Classification on ImageNet (using extra training data)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Image Classification ImageNet ResNeXt-101 32×16d Top 1 Accuracy 84.2% # 312
Number of params 194M # 894
GFLOPs 72 # 441
Image Classification ImageNet ResNeXt-101 32x48d Top 1 Accuracy 85.4% # 221
Number of params 829M # 951
GFLOPs 306 # 478
Image Classification ImageNet ResNeXt-101 32x32d Top 1 Accuracy 85.1% # 244
Number of params 466M # 932
GFLOPs 174 # 467
Image Classification ImageNet ResNeXt-101 32x8d Top 1 Accuracy 82.2% # 508
Number of params 88M # 831

Methods