Dog and Cat Classification CNN

A repo for building a CNN from scratch to classify Cats and Dogs in a picture with varying qualities

Dataset

Dogs vs. Cats Image Classification Dataset
Link

Project Proposal

Project ideas and methodology
Link

Project Backlog

Tasks monitoring and milestones
Link

Project Slide

Presentation Slides
Link

Set up

Start a virtual environment (this requires you to have a python 3.11 installed):

py -3.11 -m venv .venv

Activate the virtual environment (from project root):

.\.venv\Scripts\Activate.ps1

Install the required libraries:

pip install -r requirements.txt

Set up your environment variable

create a file named `.env` in project root

inside the `.env`, set up the correct variables

You should have your training and testing data folder

in the structure of

<folder name>
| - train
| - |- dog
| - |- cat
|
| - test
| - |- dog
| - |- cat

your .env should look like

input_train_data_location="<path to the whole train dataset>"
input_test_data_location="<path to the whole test dataset>"

output_train_data_location="<path to the filtered train dataset>"
output_test_data_location="<path to the filtered test dataset>"

Filter the original dataset to a smaller subset

We used file size as the filter (keeping size > 16 kB)

run

python .\preprocessing.py

you should see a filtered dataset folder being created, this would be the data for the model to train

Train the Pre-Trained model

Train the model using the training dataset,

the script will automatically split the training set into a 80%, 20% segments

and use the 20% segment as validation set.

Running the training script might take longer time if running on GPU,

I waited about 1 hour for my laptop to train on the dataset

python .\infr_EfficientNet.py

Once done, you should see a model.pth in the project root,

that is the trained weight

Use the Trained model

Test your trained efficientnet instance by running

python .\use_trained_eff_net.py

the end result should be Dog (1.0000), where 1.0000 represents the probabilty