/pytorch-deeplab-xception

DeepLab v3+ model in PyTorch

Primary LanguagePythonMIT LicenseMIT

pytorch-deeplab-xception

Introduction

This is a Pytorch implementation of DeepLab-Xception, It's based on a Modified Aligned Xception backbone. Currently, we train DeepLab V3+ from scratch, using Pascal Voc 2012 dataset.

Results

We use deeplab v3+ model trained on Pascal VOC 2012 and SBD datasets to inference these results. After 50 training epoch, our deeplab v3+ model can reach 72.7% mIoU on Pascal VOC 2012 test set.

Requirements

Python3.x (Tested with 3.5)
PyTorch (Tested with 0.4.0)
tensorboardX
opencv-python

Installation

This code is tested in Ubuntu 16.04. To use this code, please do:

  1. Clone the repo:

    git clone https://github.com/jfzhang95/pytorch-deeplab-xception.git
cd pytorch-deeplab-xception

  2. To train DeepLabV3+, please do:

    python train.py

TODO

  • Basic deeplab v3+ model
  • Training deeplab v3+ on SBD dataset
  • Results evaluation on Pascal VOC 2012 test set
  • Deeplab v3+ model using resnet as backbone
  • Training deeplab v3+ on other datasets