This repo is the tensorflow implementation of Discriminative Feature Network (DFN)
The original paper for DFN can be found at https://arxiv.org/abs/1804.09337.
Most existing methods of semantic segmentation still suffer from two aspects of challenges: intra-class inconsistency and inter-class indistinction. To tackle these two problems, we propose a Discriminative Feature Network (DFN), which contains two sub-networks: Smooth Network and Border Network. Specifically, to handle the intra-class inconsistency problem, we specially design a Smooth Network with Channel Attention Block and global average pooling to select the more discriminative features. Furthermore, we propose a Border Network to make the bilateral features of boundary distinguishable with deep semantic boundary supervision.
This repo of code is written using Tensorflow. Please install all required packages before using this code.
pip install -r requirements.txtThe project can only solve problems of binary segmentation at present. So you can only collect a dataset with labels of binary segmentation and split into 3 directories labeled 'train', 'val' and 'test', each of which has 2 subfolders named 'main' and 'segmentation', while 'main' should store original images and 'segemntation' should store segmentation images. The name of a image must be same as that of its corresponding segmentation image.
python data_augment.py --dir data/trainpython main.py --batch_size 1python main.py --batch_size 1 --is_training FalseThe results would be saved in the folder test-outputs.
python evaluation.py --gt_dir data/test/segmentation --pred_dir test-outputs --result_txt results.txtThe IOU results would be written in the results.txt.
Please direct any questions or comments to me; I am happy to help in any way I can. You can email me directly at 20164228014@stu.suda.edu.cn.