|
|
--- |
|
|
license: mit |
|
|
language: en |
|
|
tags: |
|
|
- pytorch |
|
|
- image-classification |
|
|
- computer-vision |
|
|
- pothole-detection |
|
|
--- |
|
|
|
|
|
# 🧠 Pothole Detection Model (ResNet18) |
|
|
|
|
|
This repository contains a fine-tuned **ResNet18** model for classifying whether a road image contains a **pothole** or not. |
|
|
The model was developed as part of the **Zindi MIIA Pothole Image Classification Challenge**, focused on improving road safety in South Africa. |
|
|
|
|
|
## 📚 Project Overview |
|
|
- **Task:** Binary image classification (`pothole` / `no_pothole`) |
|
|
- **Framework:** PyTorch + torchvision |
|
|
- **Model:** ResNet18 (pretrained on ImageNet, fine-tuned) |
|
|
- **Evaluation Metric:** AUC (Area Under the Curve) |
|
|
- **Dataset Source:** [rupesh002/Patholes_Dataset](https://huggingface.co/datasets/rupesh002/Patholes_Dataset) |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔗 Full Project Code |
|
|
The complete project (training, inference, and notebook) is available on GitHub: |
|
|
👉 [MIIA Pothole Image Classification on GitHub](https://github.com/Rupeshbhardwaj002/MIIA_Pothole_Image_Classification_SouthAfrica) |
|
|
|
|
|
It includes: |
|
|
- Data preprocessing |
|
|
- Model training pipeline (`train.py`) |
|
|
- Prediction and submission script (`predict.py`) |
|
|
- Notebook used for Zindi competition submission |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚙️ How to Load This Model |
|
|
|
|
|
```python |
|
|
from huggingface_hub import hf_hub_download |
|
|
import joblib |
|
|
|
|
|
model_path = hf_hub_download( |
|
|
repo_id="rupesh002/pothole_detection_model", |
|
|
filename="model.pkl", |
|
|
repo_type="model" |
|
|
) |
|
|
|
|
|
model = joblib.load(model_path) |
|
|
--- |
|
|
🏷️ Author |
|
|
|
|
|
Rupesh Bhardwaj |
|
|
Machine Learning Enthusiast |
