Prerequisites

Set up virtual environment

python3 -m venv ./venv
source ./venv/bin/activate
pip install -r requirements.txt

Get the datasets

1. Download the first dataset and place it in ./data/d1.zip
2. Download the second dataset and place it in ./data/d2.zip
3. Download the third dataset and place it in ./data/d3.zip
4. Run the following commands from the project root (./):

unzip ./data/d1.zip -d ./data/d1
unzip ./data/d2.zip -d ./data/d2
unzip ./data/d3.zip -d ./data/d3

---

Project Structure

Data Cleaning and Preprocessing

• preprocess.ipynb — Processes and combines the three datasets into one and stores it in CSV format as trash.csv.
• feature_extraction.ipynb — Uses a ResNet (CNN) model to extract features from each image in trash.csv and stores them in final.csv.

---

Models

• random_forest_model.ipynb — Uses a Random Forest classifier and achieves 99.17% accuracy.
• xgboost_model.ipynb — Uses the XGBoost algorithm and achieves 99.61% accuracy.
• lightgbm_model.ipynb — Uses the LightGBM algorithm and achieves 99.63% accuracy.

---

Note: Since it might take several hours to generate final.csv in feature_extraction.ipynb, you can directly download it from Google Drive and place it in ./data/final.csv.

---

Inference

Now you can test the model on your own images! To do this:

1. Download or provide an image.
2. Run the following command with the correct path to your image:

   python3 inference.py path/to/image