jianlizh429 / dog_vs_cat

Dog VS Cat

Kaggle competition: Dogs vs. Cats Redux: Kernels Edition

Usage

Format dataset

Format the dataset to the architecture that fit keras

data/
    train/
        dogs/
            dog001.jpg
            dog002.jpg
            ...
        cats/
            cat001.jpg
            cat002.jpg
            ...
    validation/
        dogs/
            dog001.jpg
            dog002.jpg
            ...
        cats/
            cat001.jpg
            cat002.jpg
            ...

$dog_vs_cat_dataset_dir: path of train directory in the dog vs cat dataset downloaded from kaggle, after you unzip the all.zip

$target_dir: the directory where you want to put your output

python train/format_dataset.py $dog_vs_cat_dataset_dir $target_dir --val_ratio 0.1

training

• train ResNet50

  python train/train_resnet50.py $train_dataset_dir $validation_dataset_dir --batch_size 256 --epochs 100 --lr 0.001

• train ResNet50 with center loss

  python train/train_resnet50_with_center_loss.py $train_dataset_dir $validation_dataset_dir --batch_size 256 --epochs 100 --lr 0.001

• train InceptionResNetV2 with center loss

  python train/train_inception_resnet_v2.py $train_dataset_dir $validation_dataset_dir --batch_size 256 --epochs 100 --lr 0.001

evaluate

$test_dataset_dir: path of test dataset directory of kaggle dog vs cat dataset

python train/evaluate.py weights_file_path $test_dataset_dir --out_dir $out_dir