Rethinking ImageNet Pre-training

ICCV 2019 Kaiming HeRoss GirshickPiotr Dollár

We report competitive results on object detection and instance segmentation on the COCO dataset using standard models trained from random initialization. The results are no worse than their ImageNet pre-training counterparts even when using the hyper-parameters of the baseline system (Mask R-CNN) that were optimized for fine-tuning pre-trained models, with the sole exception of increasing the number of training iterations so the randomly initialized models may converge... (read more)

PDF Abstract

Evaluation Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK COMPARE
Object Detection COCO minival Mask R-CNN (ResNet-101-FPN, GN, Cascade) box AP 47.4 # 3
Object Detection COCO minival Mask R-CNN (ResNeXt-152-FPN, cascade) box AP 48.6 # 1
Object Detection COCO minival Mask R-CNN (ResNeXt-152-FPN, cascade) AP50 66.8 # 4
Object Detection COCO minival Mask R-CNN (ResNeXt-152-FPN, cascade) AP75 52.9 # 1
Object Detection COCO minival Mask R-CNN (ResNeXt-152-FPN) box AP 46.4 # 6
Object Detection COCO minival Mask R-CNN (ResNeXt-152-FPN) AP50 67.1 # 2
Object Detection COCO minival Mask R-CNN (ResNeXt-152-FPN) AP75 51.1 # 3