README.md

📂 This repository is now archived as a finished project.

Butterfly Classification

This is a simple repository of my final project for iD Tech's Artificial Intelligence and Machine Learning with NVIDIA. The repository contains the files needed to generate a model for butterfly classification, as well as code for exporting and validating the model.

Note: Most of these commands need to be run on a Jetson Nano; however, you can train the model on a Google Colab notebook.

Getting Started

To get started, first download the dataset from here and unzip it. This dataset contains images of various butterfly species and their corresponding names.

Then, create a data directory in the root of the repository with mkdir data, and place the dataset in the data folder with the following structure:

data
├─── test
│   ├─── ADONIS
│   ├───   ├─── 1.jpg
│   ├───   ├─── 2.jpg
│   ├───   ├─── ...
├─── train
│   ├─── ADONIS
│   ├───   ├─── 01.jpg
│   ├───   ├─── 02.jpg
│   ├───   ├─── ...
├─── val
│   ├─── ADONIS
│   ├───   ├─── 01.jpg
│   ├───   ├─── 02.jpg
│   ├───   ├─── ...
├─── labels.txt

After downloading the dataset, you can start training the model by running the following command:

python3 train.py --model-dir=model --batch-size=32 --workers=4 --epochs=30 data

Exporting the Model

The model can be exported to a .onnx file by running the following command:

python3 onnx_export.py --model-dir=model

To validate the model, run the following command:

python3 onnx_validate.py --model=model

Testing the Model

To test the accuracy of the model, download the 6 test images from here and place them in the test folder.

Then, run the following commands:

imagenet.py --model=model/resnet18.onnx --input_blob=input_0 --output_blob=output_0 --labels=data/labels.txt test/<id>.jpg

where <id> is the id of the image.