Skip to content

Latest commit

 

History

History
76 lines (67 loc) · 3.37 KB

README.md

File metadata and controls

76 lines (67 loc) · 3.37 KB
Raw

DISE-Domain-Invariant-Structure-Extraction

Pytorch Implementation of the paper All about Structure: Adapting Structural Information across Domains for Boosting Semantic Segmentation, CVPR 2019.

Introduction video Paper (ArXiv) Project Page

Paper

All about Structure: Adapting Structural Information across Domains for Boosting Semantic Segmentation
Wei-Lun Chang*, Hui-Po Wang*, Wen-Hsiao Peng, Wei-Chen Chiu (*contribute equally)
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019.

@inproceedings{chang2019all,
 title={All about Structure: Adapting Structural Information across Domains for Boosting Semantic Segmentation},
 author={Chang, Wei-Lun and Wang, Hui-Po and Peng, Wen-Hsiao and Chiu, Wei-Chen},
 booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
 year={2019}
}

Example Results

prediction_results.png

Quantitative Results

performance.png

Prerequisite

  • Pytorch 0.3.1
  • Nvidia GPU with at least 16 GB memory

Installation

git clone https://github.com/a514514772/DISE-Domain-Invariant-Structure-Extraction.git

Datasets

  1. Download the GTA5 Dataset as the source domain and unzip it to /data
  2. Download the Cityscapes Dataset as the target domain and unzip it to /data

The structure of /data may look like this:

├── data
    ├── Cityscapes
    │   ├── gtFine
    │   └── leftImg8bit
    ├── GTA5
        ├── images
        └── labels

Usage

Pretrained Weights

Google drive

Example Training Script: GTA5 to Cityscapes

python train_dise_gta2city.py --gta5_data_path /data/GTA5 --city_data_path /data/Cityscapes

Example Testing Script:

Note that, to test performance on the testing set, we provide scripts to generate 1024x2048 outputs which are compatible with the testing server.

python evaluate.py ./weights --city_data_path /data/Cityscapes

More options

python train_dise_gta2city.py  -h
usage: train_dise_gta2city.py [-h] [--dump_logs DUMP_LOGS] [--log_dir LOG_DIR] [--gen_img_dir GEN_IMG_DIR]
                              [--gta5_data_path GTA5_DATA_PATH] [--city_data_path CITY_DATA_PATH]
                              [--data_list_path_gta5 DATA_LIST_PATH_GTA5]
                              [--data_list_path_city_img DATA_LIST_PATH_CITY_IMG]
                              [--data_list_path_city_lbl DATA_LIST_PATH_CITY_LBL]
                              [--data_list_path_val_img DATA_LIST_PATH_VAL_IMG]
                              [--data_list_path_val_lbl DATA_LIST_PATH_VAL_LBL]
                              [--cuda_device_id CUDA_DEVICE_ID [CUDA_DEVICE_ID ...]]

Domain Invariant Structure Extraction (DISE) for unsupervised domain adaptation for semantic segmentation

Acknowledgement

We implement this project heavily based on AdaptSeg proposed by Tsai et el..