puravparab / animal_classifier Goto Github PK

This is image classification model trained using convolutional neural networks to classify animal images.

Clone repository

git clone https://github.com/puravparab/animal_classifier.git

Run setup.sh

./setup.sh

Activate conda environment

conda activate <env name>

Install python packages to conda (if required):

conda install -y jupyter matplotlib numpy
conda install pytorch torchvision -c pytorch

Run jupyter notebook

cd notebooks
jupyter notebook

Download the dataset from kaggle into /datasets/animals directory

• 117 classes: https://www.kaggle.com/datasets/npurav/animals
• 151 classes: https://www.kaggle.com/datasets/npurav/animal-classification-dataset

1. CNN Large

• Params: 22,021,469
• Training Set: 138,420 images
• Number of classes: 117
• Accuracy: 98%
• Test Loss: 0.0415
• Epochs: 20

2. CNN Medium

• Params: 4,577,121
• Training Set: 138,420 images
• Number of classes: 117
• Accuracy: 97%
• Test Loss: 0.090
• Epochs trained: 20

3. CNN Small

• Params: 449,125
• Training Set: 138,420 images
• Number of classes: 117
• Accuracy: 86%
• Test Loss: 0.4769
• Epochs trained: 40