AustNet-Inharmonious-Region-Localization

This is the official code of the paper:

Inharmonious Region Localization with Auxiliary Style Feature
Penghao Wu, Li Niu, Liqing Zhang
arXiv Paper, BMVC 2022

Online Demo

Try this online demo for image composition (object insertion) built upon libcom toolbox and have fun!

![]

Install

Clone this repo and build the environment

git clone https://github.com/bcmi/AustNet-Inharmonious-Region-Localization.git
cd AustNet-Inharmonious-Region-Localization
conda env create -f environment.yml --name Austnet
conda activate Austnet

Download the semantic segmentation network model weight through link Dropbox or Baidu Yun with code pfpy. Put the model weight in the HRNet-Semantic-Segmentation-HRNet-OCR folder.

Datset

Please refer to DIRL to download the iHarmoney4 dataset.

Training

To train AustNet, run

python train_austnet.py --dataset_root PATH_OF_THE_DATASET --logdir austnet_training_log --gpus NUMBER_OF_GPUS

To train AustNet_S, run

python train_austnet_s.py --dataset_root PATH_OF_THE_DATASET --logdir austnet_s_training_log --gpus NUMBER_OF_GPUS

Pretrained Model

Model	Google Drive Link	Baidu Yun Link
Austnet	Dropbox	Baidu Yun code: m8ku
Austnet_s	Dropbox	Baidu Yun code: jrdi

Evaluation

To evaluate AustNet, run

python test_austnet.py --dataset_root PATH_OF_THE_DATASET --ckpt MODEL_WEIGHT_PATH

To evaluate AustNet_S, run

python test_austnet_s.py --dataset_root PATH_OF_THE_DATASET --ckpt MODEL_WEIGHT_PATH

Citation

If you find our work or code helpful, please cite:

@inproceedings{Wu2022Inharmonious,
  title={Inharmonious Region Localization with Auxiliary Style Feature},
  author={Penghao Wu and Li Niu and Liqing Zhang},
  booktitle={BMVC},
  year={2022}
}

Acknowledgement

Our code is based on repositories:

• DIRL
• HRNet-OCR