XDenseQNet

KO | EN

gitlite — search

#python #react #golang #go #android #machine-learning #rust #docker #nodejs #cli #kubernetes #awesome-list #pytorch #mcp

(★ 2)

XDenseQNet: An Explainable Quantum-Enhanced DenseNet121 for Monkeypox Skin Lesion Classification

XDenseQNet 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Monkeypox classification

> A hybrid quantum-classical deep learning framework combining DenseNet121, CBAM attention, and a parameterised quantum circuit for 4-class skin image classification on the MSID dataset.

[![License](https://img.shields.io/badge/license-MIT-blue)](https://raw.githubusercontent.com/umairinayat/XDenseQNet/main/LICENSE)
[![Python](https://img.shields.io/badge/python-3.8%2B-blue)]()
[![PyTorch](https://img.shields.io/badge/PyTorch-2.1%2B-ee4c2c)]()
[![PennyLane](https://img.shields.io/badge/PennyLane-0.36%2B-4CAF50)]()

---

## Abstract

---

## Architecture

**Proposed Method**: Input Image &rarr; DenseNet121 Backbone &rarr; CBAM Attention &rarr; Quantum Feature Extraction (4 qubits, 2 layers) &rarr; Feature Concatenation &rarr; Dense Classifier &rarr; Prediction

![XDenseQNet Architecture](https://raw.githubusercontent.com/umairinayat/XDenseQNet/main/utils/docs/XDenseQNet.jpg)

---

## Results

### Proposed Method vs. Backbone Comparisons (4Q-2L Configuration)

| Model | Accuracy | F1 | Precision | Recall | ROC-AUC | Inference (ms) |
|-------|----------|-----|-----------|--------|---------|----------------|
| **XDenseQNet (Ours)** | **95.83%** | **0.9470** | **0.9468** | **0.9531** | **0.9945** | **9.85** |
| ConvNeXt-Tiny + 4Q-2L | 94.17% | 0.9350 | — | — | 0.9937 | — |
| EfficientNet-B0 + 4Q-2L | 92.50% | 0.9140 | — | — | 0.9868 | — |
| MobileNet-V2 + 4Q-2L | 92.50% | 0.9066 | — | — | 0.9945 | — |
| ResNet50 + 4Q-2L | 90.83% | 0.8902 | — | — | 0.9917 | — |
| Swin-Tiny + 4Q-2L | 90.00% | 0.8732 | — | — | 0.9705 | — |
| ResNet18 + 4Q-2L | 88.33% | 0.8571 | — | — | 0.9825 | — |
| VGG16 + 4Q-2L | 87.50% | 0.8503 | — | — | 0.9745 | — |

### Ablation Study

| Variant | Description | Accuracy | F1 | ROC-AUC |
|---------|-------------|----------|-----|---------|
| **Full (Ours)** | DenseNet121 + CBAM + 4Q-2L QNN | **95.83%** | **0.9470** | **0.9945** |
| A1 | DenseNet121 + CBAM + Dense Head (no QNN) | 90.00% | 0.8849 | 0.9792 |
| A2 (SVM-RBF) | DenseNet121 + CBAM features → SVM | 82.50% | 0.7747 | 0.9613 |
| A3 | DenseNet121 + Standard Head (no CBAM, no QNN) | 88.33% | 0.8532 | 0.9739 |

*Tested on [MSID](https://www.kaggle.com/datasets/dipuiucse/monkeypoxskinimagedataset). See `results/` for full tables and figures.*

---

## Quick Start

### 1. Clone & Install

```bash
git clone https://github.com/umairinayat/XDenseQNet.git
cd XDenseQNet
pip install -r requirements.txt
```

### 2. Prepare Data

```bash
# Download MSID from Kaggle:
# https://www.kaggle.com/datasets/dipuiucse/monkeypoxskinimagedataset
# Place the extracted folders in data/Monkeypox Skin Image Dataset

# Run balanced augmentation + train/val/test split:
python train.py --config configs/proposed.yaml --prepare-data --skip-phase1 --device cpu
```

See [`data/README.md`](data/README.md) for detailed dataset setup instructions.

### 3. Train

```bash
# Full two-phase training (proposed method)
python train.py --config configs/proposed.yaml

# Train a specific baseline
python train.py --config configs/baselines/resnet50.yaml

# Skip Phase 1 if backbone is already fine-tuned
python train.py --config configs/proposed.yaml --skip-phase1
```

### 4. Evaluate

```bash
python evaluate.py --checkpoint checkpoints/best.pth --config configs/proposed.yaml
```

This generates confusion matrices, ROC curves, per-class bar charts, and CSV metrics in `results/`.

---

## Experiments

All experiments are reproducible via the notebooks in `notebooks/`:

| Notebook | Description |
|----------|-------------|
| [`01_proposed_method.ipynb`](notebooks/01_proposed_method.ipynb) | Trains 8 backbones × 2 quantum configs (4Q-2L, 6Q-3L), produces comparison tables |
| [`02_ablation_study.ipynb`](notebooks/02_ablation_study.ipynb) | Ablation A1 (no QNN), A2 (classical ML), A3 (no CBAM/QNN), with visualisations |

---

## Requirements

See [`requirements.txt`](requirements.txt). Key dependencies:

- Python 3.8+
- PyTorch 2.1+
- PennyLane 0.36+ (quantum circuit simulation)
- torchvision 0.16+
- timm 1.0+ (ConvNeXt, Swin Transformer backbones)
- Albumentations 1.3+ (data augmentation)
- scikit-learn 1.3+

---

## Citation

If you use this code in your research, please cite:

```bibtex
@article{tariq2026xdenseqnet,
  title   = {XDenseQNet: Hybrid Quantum Neural Network for Skin Lesion Classification},
  author  = {Tariq Shehroz, Inayat Umair, Zulfiqar Fatima, Raza Rehan and Arif Muhammad},
  year    = {2026},
  url     = {https://github.com/umairinayat/XDenseQNet}
}
```

---

## License

This project is licensed under the MIT License — see [LICENSE](LICENSE) for details.

---

## Acknowledgements

- [MSID Dataset](https://www.kaggle.com/datasets/dipuiucse/monkeypoxskinimagedataset) — Monkeypox Skin Image Dataset
- [PennyLane](https://pennylane.ai/) — Quantum machine learning framework
- [PyTorch](https://pytorch.org/) — Deep learning framework
- [timm](https://github.com/huggingface/pytorch-image-models) — PyTorch Image Models

XDenseQNet

XDenseQNet 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

XDenseQNet

XDenseQNet 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README