This repository contains the official implementation of the paper YOEO.
pip3 install -e .git clone https://github.com/zyxunh/Entity_for_entity_erasure.git third_party/Entity
cd third_party/Entity/Entityv2
bash init.sh
cd -download Mask2Former_hornet_3x_576d0b.pth from https://huggingface.co/datasets/qqlu1992/Adobe_EntitySeg/tree/main/CropFormer_model/Entity_Segmentation/Mask2Former_hornet_3x, then modify third_party/Entity/Entityv2/CropFormer/configs/entityv2/entity_segmentation/mask2former_hornet_3x.yaml WEIGHTS to your model path.
Download dataset from https://pan.baidu.com/s/1vEXvw60QbUaR-WB0lxbvCA code: 74rh
cd unhcv/projects/rlhf
# stage1
bash train_rlhf.sh 1 --train_batch_size 10 \
--gradient_accumulation_steps 100 --save_steps 400 --train_steps 800 \
--dataset_config config/dataset_config/dataset.py --model_config config/model_config/yoeo1.yml
# stage2
bash train_rlhf.sh 1 --train_batch_size 7 \
--save_steps 5000 --train_steps 15000 \
--dataset_config config/dataset_config/dataset.py --model_config config/model_config/yoeo2.yml
# stage3
bash train_rlhf.sh 1 --train_batch_size 7 \
--save_steps 1000 --train_steps 3600 \
--dataset_config config/dataset_config/dataset_with_pair0.py --model_config config/model_config/yoeo3.ymlDownload model from https://huggingface.co/unhzyx/yoeo/tree/main
python3 unhcv/projects/rlhf/test.pyThis project makes use of the following open-source repositories:
We sincerely thank the authors for open-sourcing their valuable work.
If you use this codebase or models in your research, please consider cite .
@inproceedings{YOEO,
title={You Only Erase Once: Erasing Anything without Bringing Unexpected Content},
author={Zhu, Yixing and Zhang, Qing and Xu, Wenju and Zheng, Wei-Shi},
booktitle={CVPR},
year={2026}
}