README.md

---
license: apache-2.0
language:
- en
datasets:
- ILSVRC/imagenet-1k
---
# Model Card for Model ID
VIT-MAE-r is a fine-tuned version of MAE for image reconstuction. We release a version fine-tuned from [MAE-Large](https://huggingface.co/facebook/vit-mae-large)

## Model Details

VIT-MAE-r is already converted to hf format and should be able to be used directly by `from_pretrained` method.

### Model Sources

<!-- Provide the basic links for the model. -->

- **Repository:** [LM4LV](https://github.com/bytetriper/LM4LV)
- **Paper:** [LM4LV: A Frozen Large Language Model for Low-level Vision Tasks](https://arxiv.org/abs/2405.15734v1)
- **source model**: [MAE-Large](https://huggingface.co/facebook/vit-mae-large)

## How to Get Started with the Model

Use the code below to get started with the model.

```python
from transformers import AutoImageProcessor, AutoModelForPreTraining
model = AutoModelForPreTraining.from_pretrained("bytetriper/vit-mae-r")
```


## Evaluation

This model achieves a rFID on ImageNet val set of 1.24, evaluated using the standard tensorflow tool provided by [Guided-Diffusion](https://github.com/openai/guided-diffusion/tree/main/evaluations)

## Citation

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

**BibTeX:**

@article{zheng2024lm4lv,
  title={LM4LV: A Frozen Large Language Model for Low-level Vision Tasks},
  author={Zheng, Boyang and Gu, Jinjin and Li, Shijun and Dong, Chao},
  journal={arXiv preprint arXiv:2405.15734},
  year={2024}
}



## Model Card Authors

Boyang Zheng

## Model Card Contact

[email protected]