MTCA-Net

A multi-task cascaded analysis network (MTCA-Net) for real-time tracking and segmentating sperm under high-resolution conditions.

📄 Corresponding Publication

This repository contains the implementation of the method described in our paper:

MTCA-Net: Multi-Task Cascade Analysis Network for Real-Time Sperm Quality Analysis

🛠 Environment Setup

This project requires Python 3.9 and PyTorch 2.0.1. Follow these steps to set up the environment:

1. Clone Repository

git clone https://github.com/Lijiajin0719/MTCA-Net.git
cd MTCA-Net

2. Create Conda Environment

conda create -n MTCA-Net python=3.9
conda activate MTCA-Net

3. Install Dependencies

pip install -r requirements.txt

📊 Dataset

1.Download datasets from SHDet, SHSeg and SHSegHR

2.Extract files to the datasets directory

🏋️ MTCA-Net train/test

1. Train

(1) Train Detection Module

python train_detect.py

To see more intermediate results, check out ./runs/detect/AMF-YOLO....

(2) Train Segmentation Module

python train_seg.py --batch_size 32 --epochs 300 --val_interval 5 --save_interval 50

To see more intermediate results, check out ./seg/run/Effusion_U2Net....

2. Test

(1) Test Detection Module

python test_detect.py

The test results will be saved to file here: ./runs/detect/test....

(2) Test Segmentation Module

python test_seg.py --model_path seg/run/EffiFusion_U2Net_.../weights/best_model.pth --ap50_threshold 0.5

The test results will be saved to file here: ./seg/run/EffiFusion_U2Net_test....

(3) Test MTCA-Net

python test_MTCA.py --detect_model_path runs/detect/AMF-YOLO/weights/best.pt --seg_model_path seg/run/EffiFusion_U2Net_test.../weights/best_model.pth --conf_threshold 0.6 --seg_threshold 0.5

The test results will be saved to file here: ./MTCA-Net/run/....

📧 Contact

For any questions regarding the paper or this implementation, please feel free to contact the authors.

📩 Email: [email protected]

📚 Acknowledgements

Our codebase is built with references to the following open-source projects:

• Ultralytics YOLO: The most popular real-time object detection model repository.

We sincerely appreciate the authors for open-sourcing their valuable work.

---

🌟 We appreciate your interest in our work!