/cs445_finalproject

Primary LanguagePython

Based on the paper Learning in the Frequency Domain

Packages:

Datasets:

File Overview:

  • conf

    • conf.yaml: stores hyperparameters and variables for training. Any value specified in conf.yaml can be replaced at runtime by specifying the new value as follows

      python train.py {parameter}={new value} where parameter is the parameter you want to replace and new value is the new value

  • datasets
    • calculate_mean_std.py: Computes the mean and standard deviation in the frequency domain
    • create_dataset.py: Creates datasets based on PyTorch's dataset format. Dataset classes include
      • TinyImagenetTrain
      • TinyImagnetTest
      • TinyImagenetTrainRGB
      • TinyImagenetTestRGB
      • ImagenetteTrain
      • ImagenetteTest
    • make_train_val_split.py: Splits training set into train and validation set and stores the values in JSON files
    • transforms.py: Contains transforms that operate in the frequency domain
  • models
    • resnet.py: Contains default source code for ResNet from PyTorch and modifications that allow data in the frequency domain to be used
  • avgs.npy: numpy array that contains (y channel average,cb channel average, cr channel average) after the discrete cosine transform for Tiny-Imagenet
  • avgs_imagenette.npy: numpy array that contains (y channel average,cb channel average, cr channel average) after the discrete cosine transform for Imagenette 160
  • avgs_imagenette_320: numpy array that contains (y channel average,cb channel average, cr channel average) after the discrete cosine transform for Imagenette 320
  • stds.npy: numpy array that contains (y channel average standard deviation, cb channel average standard deviation, cr channel average standard deviation) after the discrete cosine transform for Tiny-Imagenet
  • stds_imagenette.npy: numpy array that contains (y channel average standard deviation, cb channel average standard deviation, cr channel average standard deviation) after the discrete cosine transform for Imagenette-160
  • stds_imagenette_320.npy: stds.npy: numpy array that contains (y channel average standard deviation, cb channel average standard deviation, cr channel average standard deviation) after the discrete cosine transform for Imagenette-320
  • train.py: main training file for images in the frequency domain. To run: python train.py
  • train_list.json: Dictionary that contains the classes and the images from those classes that are part of the training set for Tiny-Imagenet
  • train_list_imagenette.json: Dictionary that contains the classes and images from those classes that are part of the training set for Imagnette (160 and 320)
  • val_list.json: Dictionary that contains the classes and the images from those classes that are part of the validation set for Tiny-Imagenet
  • val_list_imagenette.json: Dictionary that contains the classes and the images from those classes that are part of the validation set ofr Imagenette(160 and 320)
  • train_rgb.py: main training file for images in the RGB domain. To run python train_rgb.py

Sources: