KAGGLE DOG VS CAT

Script

Primary files

build_dog_vs_cat.py

• info:
  • splits and builds dataset in HDF5 format
• arguments:
  • none

train_alexnet.py

• info:
  • train AlexNet model on dataset generated by build_dog_vs_cat.py
• arguments:
  • none

evaluate.py

• info:
  • evaluates AlexNet model on test data split generated by build_dog_vs_cat.py
• arguments:
  • none

Alternative files

alt_build_dog_vs_cat.py

• info:
  • rebuilds dataset structure to one suitable for tensorflow.keras.preprocessing.image.ImageDataGenerator
• arguments:
  • none
• alternative:
  • build_dog_vs_cat.py

alt_train_alexnet.py

• info:
  • train AlexNet model on dataset generated by alt_build_dog_vs_cat.py
• arguments:
  • none
• alternative:
  • train_alexent.py

alt_evaluate.py

• info:
  • evaluates AlexNet model on test data split generated by alt_build_dog_vs_cat.py
• arguments:
  • none
• alternative:
  • evaluate.py

Feature Extraction

extract_features.py

• info:
  • builds new dataset with image features extracted with pretrained ResNet50 model
• arguments:
  • —dataset[required]: path to the raw image dataset
  • —trainpath[_optional] : full path to output train split of extracted features
  • —valpath[_optional] : full path to output train-validation split of extracted features
  • —testpath[_optional] : full path to output test split of extracted features
  • —test-size[optional]: float value size of non-training(test + train-validation) dataset split
  • —batch-size[optional] : size of feature and label samples to train model per each steps

train_model.py

• info:
  • trains and evaluates BoboNet model on extracted features
• arguments:
  • —model[required]: path to output trained model
  • —classes[optional] : number of uniques classes
  • —epochs[optional]: number of training epochs
  • —batch-size[optional] : number of feature and label samples to train model per each steps
  • —train-db[optional] : path to train dataset(HDF5 format) split
  • —test-db[optional] : path to test dataset(HDF5 format) split
  • —val-db[optional] : path to train-validation dataset(HDF5 format) split