AOGNets: Deep AND-OR Grammar Network for Visual Recognition

This repository contains the code and trained models of:

Xilai Li, Tianfu Wu*, Xi Song, and Hamid Krim, AOGNets: "Deep AND-OR Grammar Networks for Visual Recognition". (arXiv:1711.05847)

Install MXNet

./code_sync_compile.sh

How to Train

imagenet

python train_aognet.py --cfg aognet/cfgs/aog_imagenet.yaml

cifar10

python train_aognet.py --cfg aognet/cfgs/aog_cifar10.yaml

cifar100

python train_aognet.py --cfg aognet/cfgs/aog_cifar100.yaml

Train Results

Current best testing error on cifar10 and cifar100

'+' means standard data augmentation with only random crop and mirroring.