Dogs_vs_Cats_Pytorch

Introduction

• This repository is for kaggle Dogs vs. Cats match, but you can utilize this code to learn how to use pytorch.
• For network, I has estabilished the structure containing the introduction of pre-trained models like VGG,Inceptionv3 and ResNet.
• For optimizer, only Adam is illustrated in my repository.

Environment

• Python 3.6
• Pytorch 0.4.0
• torchvision 0.2.0
• tensorboardX 1.5

Dataset Structure

Original

${project_dir}/datasets
    dogs_vs_cats
        train.zip
        test1.zip

Extract train and test datasets

After downloading the datasets from Kaggle website, you need to extract these two zips.(Actually, I just extract train.zip)

${project_dir}/datasets
    dogs-vs-cats
        train.zip
        test1.zip
        Dataset           # Extracted from train.zip
        test1           # Extracted from test1.zip

Final step

• In fact, the filenames of train and test datasets is in different naming conventions.
• For train one, the filename name is in cat.x.jpg or dog.x.jpg. However, x.jpg is used in test dataset.
• To do the classification task easily, I just use the train dataset(Dataset). So if you want to build a new test/val dataset. You need to run redistribution_image.py to split the train dataset into a new one and an extra test dataset.

${project_dir}/datasets
    dogs-vs-cats
        train.zip
        test1.zip
        train           # Separated from 'train' directory by run `redistribution_image.py`.
        val             # Separated from 'train' directory by run `redistribution_image.py`.
        Dataset         # Extracted from train.zip
        test1           # Extracted from test1.zip

Experimental Directory Structure

• Before training, you need to modify the directories in demo.py
• Run demo.py to start the training process. The follow directories will be created automatically.

${project_dir}
    checkpoint
        inceptionv3
            ckpt.t7
        resnet50
            ckpt.t7
        vgg
            ckpt.t7
    log(tensorboard dirctory)
        inceptionv3
        resnet50
        vgg
    model
        inceptionv3
            net_ 1.pth
            net_ 2.pth
            ...
        resnet50
            net_ 1.pth
            net_ 2.pth
            ...
        vgg
            net_ 1.pth
            net_ 2.pth
            ...
    record
        inceptionv3
            acc.txt
            log.txt
        resnet50
            acc.txt
            log.txt
        vgg
            acc.txt
            log.txt
            ...

TensorboardX

• You can walk into log\$(use_model) directory to monitor the loss. Run tensorboard --logdir . then open the browser.

Performances

Network	Test Accuracy(%)	batch_size
VGG19	96.00	4
ResNet50	96.00	32
InceptionV3	76.00	32

• I has just trained the models for 1 epochs by 'Adam'.

Attention

If the train or test accuracy is low, you can modify the optimizer code to

optimizer = torch.optim.Adam(use_model.parameters())

to let all the parameters for training.

Besides, you need to comment the following code:

for parma in use_model.feature.parameters():
    parma.requires_grad = False
for index, parma in enumerate(use_model.classifier.parameters()):
    if index == 6:
        parma.requires_grad = True