[🩺CVAMD @ ICCV2025🌺] BAMPolyp: Bi-Axial Mamba Bottleneck for Gastrointestinal Polyp Segmentation

Official PyTorch implementation of the paper
📄 BAMPolyp: Bi-Axial Mamba Bottleneck for Gastrointestinal Polyp Segmentation
🏆 Accepted at ICCV 2025, CVAMD Workshop

📍 Paper PDF (CVF Open Access): Click Here
🎴 Poster: Click Here
👨🏻‍🔬 Authors: Md. Farhadul Islam, Tashik Ahmed, Partho Chanda, Joyanta J. Mondal, Meem Arafat Manab, Sarah Zabeen & Jannatun Noor

---

🧠 Overview

BAMPolyp is a lightweight yet powerful deep learning architecture for the segmentation of gastrointestinal (GI) polyps from colonoscopic images. The core idea is to embed a Bi-Axial Mamba bottleneck into a U-Net-style segmentation pipeline to bridge local boundary precision and global contextual coherence — crucial for high-accuracy diagnosis support systems.

---

✨ Highlights

• ✅ Novel Bi-Axial Mamba Bottleneck that performs separate axial state-space mixing along both height and width axes.
• 📦 Efficient local-global context fusion using lightweight Mamba blocks.
• ⚡ Built with a pretrained EfficientNet-B0 encoder, optimized for medical image feature extraction.
• 🔁 Uses deep supervision and residual refinement for improved convergence and multi-scale learning.
• 🔬 Validated across 4 public polyp segmentation datasets: Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, and PolypGen.
• 🧮 Only 6.51M parameters and 3.13 GFLOPs, making it deployable in resource-constrained environments.

---

📊 Benchmark Results

Dataset	Dice	IoU
Kvasir-SEG	0.9380	0.8881
ClinicDB	0.9437	0.8939
ColonDB	0.9255	0.8659
PolypGen	0.8683	0.8211

🧪 BAMPolyp consistently outperforms state-of-the-art CNN/Transformer/Mamba models in segmentation performance and computational efficiency.

---

🏗️ Architecture

Architecture Summary:

• Encoder: EfficientNet-B0
• Bottleneck: Bi-Axial Mamba Block (separate width- and height-axis Mamba SSM)
• Decoder: U-Net-style upsampling with deep supervision
• Losses: Composite (BCE + Dice + IoU + Focal + Tversky + Boundary)

---

Quick Run

conda env create -f requirements.yaml # Install for conda env (preferred)
pip install -r requirements.txt # Install for python env

# activate the bampolyp environment, and then:
python bampolyp.py

If you want to run it on Google Colab, you can feel free to use the ipynb file.

📚 Citation

If you find BAMPolyp useful in your research, please cite our paper:

@InProceedings{Islam_2025_ICCV,
    author    = {Islam, Md. Farhadul and Ahmed, Tashik and Chanda, Partho and Mondal, Joyanta Jyoti and Manab, Meem Arafat and Zabeen, Sarah and Noor, Jannatun},
    title     = {BAMPolyp: Bi-Axial Mamba Bottleneck for Gastrointestinal Polyp Segmentation},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
    month     = {October},
    year      = {2025},
    pages     = {1082-1092}
}