Skip to content

chaser164/dog-image-classification

Repository files navigation

PuppyFinder: A Dog Image Classification Project

Tyler Tan & Chase Reynders

Setup

First, we recommend cloning this repository and running the ipynb file locally (i.e. not on colab), as it requires a few auxiliary files that will be easier to manage outside of Google Drive.

Next, download the archive.zip file from here and place this .zip file in the same directory that contains Tyler_&_Chase_Machine_Learning_Final_Project_implementation.ipynb. Be sure that it is named archive.zip. No need to unzip it yourself; the notebook will take care of this for you in the pre-processing step.

Next, we recommend creating a python virtual environment by running python -m venv .venv from the directory containing the .ipynb file. We suggest you do this so that the pip installs are not installed into your global python environment.

Be sure to choose the virtual environment you just created as the .ipynb file's selected environment as demonstrated here.

Before running all chunks, note the second code chunk in the notebook. use_old_model is set to True in order to give you a brief demo of one of our already-trained models (this will only take a minute or so). If you wish to train a new model (which in our experience took a few hours), please set this boolean to False.

Now, you should be able to run all of the .ipynb files's chunks. If prompted to do so, install the ipykernel packages.

Note: you likely won't run into this, but we found that installing the ipykernel packages only works when pip3 is aliased to pip.

Helpful Resources for this project:

Pytorch transfer learning approach (quite helpful): https://pytorch.org/tutorials/beginner/transfer_learning_tutorial.html https://pytorch.org/hub/pytorch_vision_resnet/

Microsoft CNN image classification learning resource: https://learn.microsoft.com/en-us/windows/ai/windows-ml/tutorials/pytorch-train-model

PyTorch classification tutorial: https://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html

About

Convolutional neural network trained on 100+ breeds that classifies with over 94% accuracy.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors