🎯 Vision Master — Scene Classification with ResNet-18

Built by Abhijeet (2025) | Powered by PyTorch + GPU Acceleration ⚡

Vision Master is a high-performance image classification system trained on the Intel Image Classification dataset.
It can accurately recognize 6 real-world scene types:

🏙️ buildings
🌲 forest
❄️ glacier
⛰️ mountain
🌊 sea
🛣️ street

This project includes training pipeline, evaluation tools, prediction script, and full visualizations — all optimized to run on your RTX 3060.

---

🚀 Features

• ✔️ Transfer Learning using ResNet-18
• ✔️ GPU-accelerated training (CUDA 12.1)
• ✔️ Clean training logs & visualizations
• ✔️ Confusion matrix + per-class accuracy
• ✔️ Ready-to-use prediction script
• ✔️ Portfolio-quality project layout
• ✔️ Easy to extend for your own dataset

---

🧠 Model Architecture

Using ResNet-18, pretrained on ImageNet and fine-tuned on 6-class scene classification.

• Optimizer: Adam
• Loss: CrossEntropy
• LR Scheduler: StepLR
• Epochs: 20
• Mixed GPU/CPU compatibility

---

📦 Project Structure

vision-master/
│── train.py               # Train the model
│── evaluate_model.py      # Evaluate accuracy & metrics
│── predict_image.py       # Predict a single custom image
│── visualize_training.py  # Plot loss/accuracy curves
│── analyze_model.py       # Confusion matrix + class accuracy
│── data/                  # Training & test dataset
│── models/                # Saved weights & graphs
│── custom/                # Your test images
│── requirements.txt       # Pip dependencies
│── README.md              # This file

---

📊 Training Visualizations

📉 Loss Curve

📈 Accuracy Curve

🧩 Confusion Matrix

🎯 Per-Class Accuracy

---

🏆 Final Model Performance

Metric	Score
Overall Accuracy	94.40%
Best Accuracy Achieved	94.40%
Epochs	20
GPU Used	NVIDIA RTX 3060
Batch Size	32

📌 Per-Class Performance

• 🏙️ Buildings — 95.65%
• 🌲 Forest — 99.58%
• ❄️ Glacier — 90.24%
• ⛰️ Mountain — 90.29%
• 🌊 Sea — 98.24%
• 🛣️ Street — 93.41%

---

⚙️ Installation

1️⃣ Clone the repository

git clone https://github.com/abhiijeetdev/vision-master.git
cd vision-master

2️⃣ Create a virtual environment

python3 -m venv venv
source venv/bin/activate

3️⃣ Install dependencies

pip install -r requirements.txt

---

🏋️ Train the Model

python train.py

Trained weights will appear in:

models/resnet18_intel_best.pth
models/resnet18_intel_last.pth

---

📊 Evaluate the Model

python eval_model.py

---

🖼️ Predict a Custom Image

Place your image inside:

custom/your_image.jpg

Then run:

python predict_image.py custom/your_image.jpg

---

👑 Author

Created by Abhijeet (2025)
Focused on AI/ML, High-Performance Vision Systems & Deep Learning Engineering.
Built entirely on Linux + RTX 3060.

---

⭐ Acknowledgements

• Intel Image Scene Dataset
• PyTorch Team
• TorchVision Models
• RTX 3060 (for going Ultra Instinct ⚡🔥)

---

If you like this project, ⭐ star it on GitHub — it boosts your profile!