https://github.com/rickydoan/deep-learning-car-damage-detection
https://github.com/rickydoan/deep-learning-car-damage-detection
Last synced: about 1 month ago
JSON representation
- Host: GitHub
- URL: https://github.com/rickydoan/deep-learning-car-damage-detection
- Owner: RickyDoan
- Created: 2025-01-09T18:07:48.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-01-10T15:13:45.000Z (over 1 year ago)
- Last Synced: 2025-02-28T06:27:28.976Z (over 1 year ago)
- Language: Jupyter Notebook
- Size: 186 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 🚗 Deep Learning - Car Damage Detection with CNNs
This repository contains an end-to-end pipeline for detecting and classifying car damages using Convolutional Neural Networks (CNNs) and pretrained models integrated with APIs and UI for deployment.
## Overview
This project aims to classify car damages into multiple categories like front crushed, rear breakages, normal conditions, etc., leveraging state-of-the-art deep learning technologies.
### Features:
- **Pretrained Models**: Transfer learning with **ResNet50**, **EfficientNet** with fine-tuning for task-specific accuracy.
- **Custom CNN**: Built from scratch and compared with pretrained methods.
- **Hyperparameter Tuning**: Automated using **Optuna**!
- **FastAPI**: An API endpoint for real-time predictions.
### Technical Highlights:
- **Data Augmentation**: Extensive transformations for robust model generalization.
- **GPU Training**: Utilized PyTorch for CUDA-accelerated training on google collab
- **Inference Pipeline**: Image preprocessing, prediction, and visualization!
### Usage:
1. **Clone Repository**
2. **Train from Scratch or Use Pretrained Weights**
### **Evaluation**:
* Confusion matrix and classification reports validate the models.
* Accuracy is around 80%