CE2P-refine# CE2P

Requirements

To install PyTorch, please refer to https://github.com/pytorch/pytorch#installation.

Compiling

Some parts of InPlace-ABN have a native CUDA implementation, which must be compiled with the following commands:

cd modules
sh build.sh
python build.py

The build.sh script assumes that the nvcc compiler is available in the current system search path. The CUDA kernels are compiled for sm_50, sm_52 and sm_61 by default. To change this (e.g. if you are using a Kepler GPU), please edit the CUDA_GENCODE variable in build.sh.

Dataset and pretrained model

Plesae download cityscapes dataset and modify the CS_PATH in job_loacl.sh

Please download imagenet pretrained resent-101, and put it into dataset folder.

Training

./job_local.sh