Pytorch Toolbox For Segmentation (V0.1)

Introduction

This repo is used for training&testing deep learning problems(segmentation). It is implemented by Python & Pytorch. (Some code from others to share, thank you for their efforts!)

Dependency

• Python3.6
• Numpy
• OpenCV3
• Pytorch
• Torchvision
• Sklearn
• tqdm

Usage

1. Preprocess

Open utils/generate_data.py and edit relevent info.

ProjectDir = "/home/dl/phoenix_lzx/torch/data"
crop_size = 320
training_data_stage1_dir = os.path.join(ProjectDir,"seaship")

img_list_1=[os.path.join(training_data_stage1_dir,'{}.jpg'.format(item)) for item in img_file_list]
label_list_1=[os.path.join(training_data_stage1_dir,'{}.png'.format(item)) for item in img_file_list]

dataset_dir=os.path.join(ProjectDir,"dataset/seaship-train")

2. Dataloader

Edit utils/seaship_loader.py and set it to fit your dataset.

3.Set model

Edit models/init.py, now this repo provide some models below.(Some code from others to share, thank you for their efforts!)

• Alexnet
• UNet
• DeepUNet
• FCN
• RefineNet
• SegNet

4.Change Train param.

Edit main.py to complete the setting. Then just RUN it.

5.Test

Edit test.py to fit your testset.

Help

This repo has just started, some problems needed to be solved. If you find it, please contact me immediately.