项目作者： AbhishekRS4

项目描述：
PSPNet implementation on Cityscapes dataset with custom classes

高级语言： Jupyter Notebook

项目主页：

项目地址: git://github.com/AbhishekRS4/psp_net.git

创建时间： 2018-11-08T16:50:18Z
项目社区：https://github.com/AbhishekRS4/psp_net
开源协议：MIT License
下载

PSPNet implementation on Cityscapes dataset

Notes

Implementation of PSPNet with ResNet-50
The original implementation uses ResNet-101 for cityscapes dataset
The image dimension used to train the model is 1024x512
15 custom classes used

Main idea

Apply pyramid pooling to feature maps of output stride 8 of input size and concatenate the output of pyramid pooling block to its input. Perform bilinear upsampling by a factor of 8

Intructions to run

To list training args use
```
python3 psp_net_train.py --help
```
To list inference args use
```
python3 psp_net_infer.py --help
```

Visualization of results - Youtube video

Reference


