This repository provides the python code for the ICRA 2020 paper:
- Xiao-Yun Zhou, Jian-Qing Zheng, Peichao Li, Guang-Zhong Yang. ACNN: a Full Resolution DCNN for Medical Image Segmentation. International Conference on Robotics and Automation 2020.
This paper proposed a method for effective atrous rate setting to achieve the largest and fully-covered receptive field with a minimum number of atrous convolutional layers. Furthermore, a new and full resolution DCNN - Atrous Convolutional Neural Network (ACNN), which incorporates cascaded atrous II-blocks, residual learning and Instance Normalization (IN) is proposed.
- Python 3.5
- Tensorflow >= 1.9
- numpy
- scipy
The code uses pre-processed .mat files as inputs, and generates inference results also in .mat files. A sample medical image scan can be found in the data folder.
Please use train_network.py for training and test_network.py for testing. Change the paths to your own data before running. The code provides ACNN, deeplab v3+ and u-net for comparison. The trained models are not provided due to the size of the file.
We thank the support of NVIDIA Corporation with the donation of the Titan Xp GPU used for this research.