This project benchmarks distributed deep learning frameworks in healthcare applications, focusing on TensorFlow (TF) 2.12 and PyTorch (Torch) 2.1. It utilizes ResNet50 and Inception V3 architectures with BreastMnist and PathMnist datasets, and employs a Deep Convolutional Generative Adversarial Network (DCGAN) with the Brats dataset. The project also explores the parameter server approach in TensorFlow and compares it with the multi-mirrored worker strategy in TF and the Distributed Data Parallel method in Torch.
ResNet50 and Inception V3 architectures BreastMnist and PathMnist datasets DCGAN with the Brats dataset #Deep Learning Frameworks TensorFlow 2.12 PyTorch 2.1
Parameter Server (TensorFlow) Multi-Mirrored Worker Strategy (TensorFlow) Distributed Data Parallel (PyTorch)
For the DCGAN implementation, please cite: "Generative Adversarial Networks in Healthcare: A Case Study on MRI Image Generation" [6]
Follow the instructions in each benchmark script to reproduce the experiments.
Feel free to contribute by opening issues or pull requests.
This project is licensed under the Universidade do Minho License.
TensorFlow team PyTorch team Professor António Luís Sousa Supervisors: Beatriz Cepa and Cláudia Brito
For any inquiries, please contact luisbranco00@outlook.pt.