neural-nets-for-images-py

The idea behind this development is that of using a previously trained pytorch model (e.g., frcnn0906.pth) to classify user provided images of cats and dogs contained in a file named: file.tar

This code can also be accessed following this link

Setup the enviroment and install dependencies

conda create -n neural-env python=3.9
conda activate neural-env
pip install -r requirements.txt

How to use the code

To execute the code, open a terminal and type:

cd app
 python neural_net.py
cd ..

Contents of the file.tar

cat.2450.jpg
cat.2451.jpg
cat.2452.jpg
cat.2453.jpg
cat.2454.jpg
cat.2455.jpg
cat.2456.jpg
cat.2457.jpg
cat.2458.jpg
cat.2459.jpg
dog.2450.jpg
dog.2451.jpg
dog.2452.jpg
dog.2453.jpg
dog.2454.jpg
dog.2455.jpg
dog.2456.jpg
dog.2457.jpg
dog.2458.jpg
dog.2459.jpg

NOTEs:

The model frcnn0906.pth was created separately with the use of a GPU infrastructure and it has an accuracy of 90% on the validation set

High-resolution images could not be tested on the Heroku Platform because the execution of the rsc468-neural APP may pass the allowed runtime

The format of the file file.tar is set to contain 20 images in total splitted evenly for images of cats and dogs (see above)

The file file.tar can be created using Unix syntax (e.g., tar -cf file.tar cat*.jpg dog*.jpg). This file can be either be uploaded if using the web APP or be saved in the main path of this repository: git@github.com:lcqsigi/neural-nets-for-images-py.git