Datasets:

Modalities:
Tabular
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:
File size: 1,818 Bytes
955f9e0
c0829ff
 
955f9e0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c0829ff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: openrail++
library_name: diffusers
dataset_info:
  features:
  - name: caption
    dtype: string
  - name: jpg_0
    dtype: binary
  - name: jpg_1
    dtype: binary
  - name: label_0
    dtype: int64
  - name: label_1
    dtype: int64
  splits:
  - name: train
    num_bytes: 2929653589
    num_examples: 1000
  download_size: 2929757570
  dataset_size: 2929653589
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# Margin-aware Preference Optimization for Aligning Diffusion Models without Reference

<div align="center">
<img src="assets/mapo_overview.png" width=750/>
</div><br>

We propose **MaPO**, a reference-free, sample-efficient, memory-friendly alignment technique for text-to-image diffusion models. For more details on the technique, please refer to our paper [here] (TODO). 

## Developed by

* Jiwoo Hong<sup>*</sup> (KAIST AI)
* Sayak Paul<sup>*</sup> (Hugging Face)
* Noah Lee (KAIST AI)
* Kashif Rasul (Hugging Face)
* James Thorne (KAIST AI)
* Jongheon Jeong (Korea University)

## Dataset

This dataset is *cartoon* split of **Pick-Style**, self-curated with [Stable Diffusion XL](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0). Using the context prompts (i.e., without stylistic specifications), we generate (1) cartoon style generation with stylistic prefix prompt and (2) normal generation with context prompt. Then, (1) is used as the chosen image, and (2) as the rejected image.

## Citation

```bibtex
@misc{todo,
    title={Margin-aware Preference Optimization for Aligning Diffusion Models without Reference}, 
    author={Jiwoo Hong and Sayak Paul and Noah Lee and Kashif Rasuland James Thorne and Jongheon Jeong},
    year={2024},
    eprint={todo},
    archivePrefix={arXiv},
    primaryClass={cs.CV,cs.LG}
}
```