VGG16-pytorch-implementation

Introduction

Here is pytorch implementation of VGG16 from scratch. It was trained on animal dataset for animal classification. It is a pratical project for basic skills in computer vision. You can down load dataset from here

Result

Due to hardware limitations, I only trained the model for approximately 28 epochs, with the highest recorded accuracy reaching 88.71%.

Figure_1

Figure_2

How to use

You can use my code as a learning material for computer vision when getting started with computer vision and wanting to code a model from scratch. Run the following command to install modules needed

pip install -r requirements.txt

The code was organized:

• Transform needed for image pre-processing: transforms.py
• Dataset was implemented in: dataset.py
• Model architecture was implemented in: vgg16_model.py
• Script to train model: train.py
• Test the model with new image: test.py
• In the notebook directory, you can also use animals_classification.ipynb and adjust some things accordingly your conditions to train on notebook or Google colab.