java>> seg_net>> 返回
项目作者： AbhishekRS4

项目描述 ：
SegNet implementation on Cityscapes dataset with custom classes
高级语言： Jupyter Notebook
项目地址: git://github.com/AbhishekRS4/seg_net.git
创建时间： 2018-11-08T14:06:27Z
项目社区：https://github.com/AbhishekRS4/seg_net
开源协议：MIT License
下载


SegNet implementation on Cityscapes dataset

Notes

  • Implementation of SegNet variant with VGG-16
  • The experiment is to use different upsampling techniques. The original implementation used Max-Unpooling.
  • The image dimension used to train the model is 1024x512
  • 15 custom classes used

Main idea

  • Unpooling mechanism used to upsample features. This uses the stored indices from corresponding pooling stage.

Intructions to run

  • To run training use
    1. python3 src/seg_net_train.py --help
  • To run inference use
    1. python3 src/seg_net_infer.py -help

Reference