j-sripad / mulitresunet-pytorch

PyTorch implementation of MultiResUNet : Rethinking the U-Net Architecture for Multimodal Biomedical Image Segmentation

MultiResUNet : Rethinking the U-Net architecture for multimodal biomedical image segmentation

This repository contains the implementation of "MultiResUNet : Rethinking the U-Net architecture for multimodal biomedical image segmentation" in pytorch.

Paper

Ibtehaz, Nabil, and M. Sohel Rahman. "MultiResUNet: Rethinking the U-Net architecture for multimodal biomedical image segmentation." Neural Networks 121 (2020): 74-87.

• Read the Paper
• Original keras(tensorflow backend) implementation

Usage

from multiresunet import MultiResUnet 

net = MultiResUnet(channels=3,filters=16,nclasses=1)





"""
Arguments : 

channels - input image channels
filters - filters to begin with (Unet)
nclasses - number of classes
"""

• Dataset - TGS Salt Identification Challenge

Results

Trained on TGS Salt Identification Challenge data for 100 epochs

mulitresunet-pytorch's Issues

The complete code package about the pytorch version

Hi! I have read the paper "MultiResUNet : Rethinking the U-Net Architecture for Multimodal Biomedical Image Segmentation" recently, and now I want to reproduce the code corresponding to the paper. I am honored to see your research on the pytorch version. Could you please provide pytorch version of the complete code package?
The gratitude is beyond words! Looking forward to your early reply!

Inplace operation cus bug

Several += operation will cause RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation.
Simply change them to a = a + b will slove the problem.