Intraoral-photo-classification

This repository provides a deep learning system for classifying intraoral dental photographs into five standard orthodontic viewpoints:

• center
• up (upper occlusal)
• down (lower occlusal)
• left
• right

The project evaluates preprocessing, data augmentation strategies, multiple image resolutions, and two versions of ResNet-18 (pretrained and trained from scratch).

1. Installation

Install dependencies:

pip install -r requirements.txt

The framework automatically uses GPU if available; otherwise CPU is used.

2. for Retraining

Run training:

python main.py

3. Inference for use the trained model

python inference.py     --patient <path_to_patient_folder>     --model <best_model.pth>     --output <results_folder>

4. Model Architecture

The classifier is based on ResNet-18 both pretrained and from Scratch.
The final fully connected layer is replaced with a 5-class output.

5. Experimental Summary

Final evaluation:

Model	Accuracy	Precision	Recall	F1‑Score	Inference Time (sec/img)
ResNet18 Pretrained	92.09%	92.15%	92.09%	92.08%	0.000732
ResNet18 Scratch	86.98%	88.06%	86.98%	86.46%	0.000760

6. Model Link

Download the model Here: https://drive.google.com/drive/u/0/folders/1LM9fIciXma-2ak9nyRqOzRrinzIH7K0S

7. Requirements

torch torchvision pillow pandas numpy shutil argparse