Slim Scissors: Segmenting Thin Object from Synthetic Background

This is the official implementation of our work Slim Scissors: Segmenting Thin Object from Synthetic Background.

Slim Scissors: Segmenting Thin Object from Synthetic Background,
Kunyang Han, Jun Hao Liew, Jiashi Feng, Huawei Tian, Yao Zhao*, Yunchao Wei
In: European Conference on Computer Vision (ECCV), 2022 [pdf] [supplementary]

Installation

Clone the repo:

git clone https://github.com/KunyangHan/SlimScissors
cd SlimScissors

Install dependencies.
```
pip install -r requirements.txt
```

Download the dataset by running the script inside data/:

cd data/
chmod +x download_dataset.sh
./download_dataset.sh
cd ..

Data folder expects the following structure:

data
├── COIFT
│   ├── images
│   ├── list
│   └── masks
├── HRSOD
│   ├── images
│   ├── list
│   └── masks
├── ThinObject5K
│   ├── images
│   ├── list
│   └── masks
└── thin_regions
    ├── coift
    │   ├── eval_mask
    │   └── gt_thin
    ├── hrsod
    │   ├── eval_mask
    │   └── gt_thin
    └── thinobject5k_test
        ├── eval_mask
        └── gt_thin

Download pretrain weights

wget https://download.pytorch.org/models/resnet18-5c106cde.pth

Training

We provide the scripts for training our models on ThinObject-5K dataset. You can start training with the following commands:

python train.py --data_root path_to_data_folder

Testing

To evaluate, simply run following commands:

python evaluation.py --ckpt path_to_pth_file --data_root path_to_data_folder --dataset target_dataset

For example

python evaluation.py --ckpt ckpt/checkpoint_epoch_29.pth --data_root ../data --dataset HRSOD

License

This project is licensed under a Creative Commons Attribution-NonCommercial 4.0 International Public License.

Citation

If you use this code, please consider citing our paper:

@InProceedings{han2022slim,
    title     = {Slim Scissors: Segmenting Thin Object from Synthetic Background},
    author    = {Han, Kunyang and Liew, Jun Hao and Feng, Jiashi and Tian, Huawei and Zhao, Yao and Wei, Yunchao},
    booktitle = {eccv},
    year      = {2022},
}