Revisiting Unreasonable Effectiveness of Data in Deep Learning Era

Benchmark Model Rank Results
image-classification-on-imagenetResNet-101 (JFT-300M Finetuning)#727Top 1 Accuracy: 79.2%
object-detection-on-cocoFaster R-CNN (ImageNet+300M)#214box mAP: 37.4AP50: 58AP75: 40.1APS: 17.5APM: 41.1APL: 51.2
pose-estimation-on-coco-test-devFaster R-CNN (ImageNet+300M)#34AP: 64.4AP50: 85.7AP75: 70.7APL: 69.8APM: 61.8
semantic-segmentation-on-pascal-voc-2012-valDeepLabv3 (ImageNet+300M)#17mIoU: 76.5%