On the Ideal Number of Groups for Isometric Gradient Propagation

Benchmark Model Rank Results
fine-grained-image-classification-on-caltechResNet-101 (ideal number of groups)Top-1 Error Rate: 22.247%
fine-grained-image-classification-on-oxford-2ResNet-101 (ideal number of groups)Accuracy: 77.076
image-classification-on-mnistMLP (ideal number of groups)Percentage error: 1.67
object-detection-on-coco-2017Faster R-CNN (ideal number of groups)AP: 40.7AP50: 61.2AP75: 44.6