[CVPR 2025 Highlight] ESC: Erasing Space Concept for Knowledge Deletion

Official PyTorch implementation for CVPR 2025 paper:

ESC: Erasing Space Concept for Knowledge Deletion

Tae-Young Lee*, Sundong Park*, Minwoo Jeon*, Hyoseok Hwang^$\dagger$ , and Gyeong-Moon Park^$\dagger$

📋 To-Do

• detailed scripts.
• upload requirements.txt.
• upload the original model checkpoints.

Environment

• Python 3.10.x
• Torch 2.1.0
• Torchvision 0.16.0
• NVIDIA GeForce RTX 3090 / A5000 / A6000
• CUDA 12.2

Getting Started

Environment

git clone git@github.com/KHU-VGI/ESC.git
cd ESC
conda create -n ESC python=3.10
conda activate ESC
pip install -r requirements.txt

Model Checkpoints

[checkpoints]

For conducting unlearning

# ESC
python unlearn.py --exp ESC_AllCNN_CIFAR10 \
                  --model_name AllCNN \
                  --method ESC \
                  --evaluation \
                  --mia \
                  --kr

# ESC-T
python unlearn.py --exp ESC_T_AllCNN_CIFAR10 \
                  --model_name AllCNN \
                  --method ESC_T \
                  --evaluation \
                  --mia \
                  --kr

BibTex

@inproceedings{lee2025esc,
    author    = {Lee, Tae-Young and Park, Sundong and Jeon, Minwoo and Hwang, Hyoseok and Park, Gyeong-Moon},
    title     = {ESC: Erasing Space Concept for Knowledge Deletion},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2025},
    pages     = {5010--5019}
}