File size: 3,182 Bytes
792f184 ac26558 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 | ---
license: mit
---
# Model Card for EAR
<!-- Provide a quick summary of what the model is/does. -->
This is an EAR (Erasing Autoregressive Models) model trained to erase specific concepts.
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** IMMC
- **Model type:** AR model
- **License:** MIT
- **Finetuned from model :** Janus-Pro
### Model Sources
<!-- Provide the basic links for the model. -->
- **Repository:** [[link](https://github.com/immc-lab/ear)]
- **Paper:** [[link](https://arxiv.org/abs/2506.20151)]
## Installation Guide
### EAR Environment
```shell
git clone https://github.com/immc-lab/ear.git
cd ear
conda create -n ear python=3.12
conda activate ear
pip install -r requirements.txt
```
### Janus-Pro Environment
Ensure that your environment can run Janus-Pro, refer to its
official [Quick Start](https://github.com/deepseek-ai/Janus) for details.
## Training Guide
After installation, follow these instructions to train EAR model for Janus-Pro.
Please run the script in `train/` after checking the file path:
```shell
python train/ear_train_church.py
```
## Generating Images with EAR
Image generation using the custom EAR model is a straightforward process. Please run the script in `infer/`.
For automated batch generation of evaluation images, utilize the following script:
```shell
python infer/infer_church.py
```
## Evaluation
You can execute the following command to evaluate the generated data. Please run the script in `eval/`.
The specific evaluation method can be found in our [paper](https://arxiv.org/pdf/2506.20151).
```shell
python eval/eval_object.py --folder_path {args.output_dir} --topk 10 --batch_size 250
```
## References
This repo is the code for the paper *EAR: Erasing Concepts from Unified Autoregressive Models*.
Thanks for the creative ideas of the pioneer researches:
- https://github.com/rohitgandikota/erasing: **Erasing Concepts from Diffusion Models**
- https://github.com/Con6924/SPM: **One-dimentional Adapter to Rule Them All: Concepts, Diffusion Models and Erasing
Applications**
- https://github.com/koushiksrivats/robust-concept-erasing: **STEREO: A Two-Stage Framework for Adversarially Robust
Concept Erasing from Text-to-Image Diffusion Models**
- https://github.com/OPTML-Group/Diffusion-MU-Attack: **To Generate or Not? Safety-Driven Unlearned Diffusion Models Are
Still Easy To Generate Unsafe Images ... For Now**
- https://github.com/deepseek-ai/Janus: **Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and
Generation**
- https://github.com/deepseek-ai/Janus: **Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model
Scaling**
## Citing our work
The preprint can be cited as follows
```bibtex
@misc{fan2025earerasingconceptsunified,
title={EAR: Erasing Concepts from Unified Autoregressive Models},
author={Haipeng Fan and Shiyuan Zhang and Baohunesitu and Zihang Guo and Huaiwen Zhang},
year={2025},
eprint={2506.20151},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2506.20151},
}
``` |