This repo contains codes for fine tuning ResNet on CUB_200_2011 datasets.
Because ResNet_SE and ResNet_ED's model files do not belong to me, so I remove them in the projects.
The ResNet models provided by torchvision are available.
CUB-200-2011 dataset has 11,788 images of 200 bird species. The project page is as follows.
Detailed information as follows:
- Images are contained in the directory data/cub200/raw/images/, with 200 subdirectories.
- Format of images.txt: <image_id> <image_name>
- Format of train_test_split.txt: <image_id> <is_training_image>
- Format of classes.txt: <class_id> <class_name>
- Format of iamge_class_labels.txt: <image_id> <class_id>
Stanford Cars datasets has 16185 images of 196 car species. The project page is as follows.
Detailed information as follows:
- Directory car_ims contains total images (both training and testing images, whose number is 16185)
- File car_nori.list contains information as follows:
git clone https://github.com/zhangyongshun/resnet_finetune_cub.git
cd base_model_finetune
#You need to modify the paths of model and data in utils/Config.py
python train.py --net_choice ResNet --model_choice 50 #ResNet50, use default setting to get the Acc reported in readme
There are some results as follows: