Tencent/tencent-ml-images
Tencent ML-Images provides a 17.6M-image multi-label dataset and a ResNet-101 model pre-trained on it, achieving 80.73% top-1 accuracy on ImageNet.

This repository releases the Tencent ML-Images dataset, the largest open-source multi-label image database with 17.6M training and 88K validation images annotated across up to 11,166 categories. It includes a ResNet-101 model pre-trained on this dataset that achieves state-of-the-art transfer learning results on ImageNet. The project provides download scripts, training procedures using TensorFlow, and checkpoints for feature extraction and fine-tuning.