MGE-CNN

Pytorch implementation of "ICCV2019-Learning a Mixture of Granularity-Specific Experts for Fine-Grained Categorization"

If you use this code, please cite our paper:

@inproceedings{zhang2019learning,
  title={Learning a Mixture of Granularity-Specific Experts for Fine-Grained Categorization},
  author={Zhang, Lianbo and Huang, Shaoli and Liu, Wei and Tao, Dacheng},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  pages={8331--8340},
  year={2019}
}

Requirement

• python 3.6
• tqdm
• yaml
• easydict
• pytorch 1.1
• pretrainedmodels
• PIL

Train

ln -s "Folder of CUB data" CUB-200-2011

python pretrain.py --config configs/cub_resnet50.yml

Inference

Pretrained model: link

python test.py --config configs/cub_resnet50.yml --model epoch_100.pth

Accuracy: 88.78 %

Reference

• Grad-CAM from jacobgil/pytorch-grad-cam