ImageNet-21K Pretraining for the Masses

22 Apr 2021  ·  Tal Ridnik, Emanuel Ben-Baruch, Asaf Noy, Lihi Zelnik-Manor ·

ImageNet-1K serves as the primary dataset for pretraining deep learning models for computer vision tasks. ImageNet-21K dataset, which is bigger and more diverse, is used less frequently for pretraining, mainly due to its complexity, low accessibility, and underestimation of its added value. This paper aims to close this gap, and make high-quality efficient pretraining on ImageNet-21K available for everyone. Via a dedicated preprocessing stage, utilization of WordNet hierarchical structure, and a novel training scheme called semantic softmax, we show that various models significantly benefit from ImageNet-21K pretraining on numerous datasets and tasks, including small mobile-oriented models. We also show that we outperform previous ImageNet-21K pretraining schemes for prominent new models like ViT and Mixer. Our proposed pretraining pipeline is efficient, accessible, and leads to SoTA reproducible results, from a publicly available dataset. The training code and pretrained models are available at:

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Image Classification CIFAR-100 ViT-B-16 (ImageNet-21K-P pretrain) Percentage correct 94.2 # 5
Multi-Label Classification MS-COCO TResNet-L-V2, (ImageNet-21K-P pretraining, resolution 640) mAP 89.8 # 11
Multi-Label Classification MS-COCO TResNet-L-V2, (ImageNet-21K-P pretraining, resolution 448) mAP 88.4 # 13
Multi-Label Classification PASCAL VOC 2007 ViT-B-16 (ImageNet-21K pretrained) mAP 93.1 # 13
Image Classification Stanford Cars TResNet-L-V2 Accuracy 96.32 # 2


No methods listed for this paper. Add relevant methods here