Update README.md
Browse files
README.md
CHANGED
@@ -1,198 +1,139 @@
|
|
1 |
---
|
|
|
2 |
library_name: diffusers
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
## Model Examination [optional]
|
135 |
-
|
136 |
-
<!-- Relevant interpretability work for the model goes here -->
|
137 |
-
|
138 |
-
[More Information Needed]
|
139 |
-
|
140 |
-
## Environmental Impact
|
141 |
-
|
142 |
-
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
|
143 |
-
|
144 |
-
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
145 |
-
|
146 |
-
- **Hardware Type:** [More Information Needed]
|
147 |
-
- **Hours used:** [More Information Needed]
|
148 |
-
- **Cloud Provider:** [More Information Needed]
|
149 |
-
- **Compute Region:** [More Information Needed]
|
150 |
-
- **Carbon Emitted:** [More Information Needed]
|
151 |
-
|
152 |
-
## Technical Specifications [optional]
|
153 |
-
|
154 |
-
### Model Architecture and Objective
|
155 |
-
|
156 |
-
[More Information Needed]
|
157 |
-
|
158 |
-
### Compute Infrastructure
|
159 |
-
|
160 |
-
[More Information Needed]
|
161 |
-
|
162 |
-
#### Hardware
|
163 |
-
|
164 |
-
[More Information Needed]
|
165 |
-
|
166 |
-
#### Software
|
167 |
-
|
168 |
-
[More Information Needed]
|
169 |
-
|
170 |
-
## Citation [optional]
|
171 |
-
|
172 |
-
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
173 |
-
|
174 |
-
**BibTeX:**
|
175 |
-
|
176 |
-
[More Information Needed]
|
177 |
-
|
178 |
-
**APA:**
|
179 |
-
|
180 |
-
[More Information Needed]
|
181 |
-
|
182 |
-
## Glossary [optional]
|
183 |
-
|
184 |
-
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
185 |
-
|
186 |
-
[More Information Needed]
|
187 |
-
|
188 |
-
## More Information [optional]
|
189 |
-
|
190 |
-
[More Information Needed]
|
191 |
-
|
192 |
-
## Model Card Authors [optional]
|
193 |
-
|
194 |
-
[More Information Needed]
|
195 |
-
|
196 |
-
## Model Card Contact
|
197 |
-
|
198 |
-
[More Information Needed]
|
|
|
1 |
---
|
2 |
+
license: openrail++
|
3 |
library_name: diffusers
|
4 |
+
tags:
|
5 |
+
- text-to-image
|
6 |
+
- text-to-image
|
7 |
+
- diffusers-training
|
8 |
+
- diffusers
|
9 |
+
- stable-diffusion-xl
|
10 |
+
- stable-diffusion-xl-diffusers
|
11 |
+
base_model: stabilityai/stable-diffusion-xl-base-1.0
|
12 |
---
|
13 |
|
14 |
+
# Margin-aware Preference Optimization for Aligning Diffusion Models without Reference
|
15 |
+
|
16 |
+
<div align="center">
|
17 |
+
<img src="https://github.com/mapo-t2i/mapo/blob/main/assets/mapo_overview.png?raw=true" width=750/>
|
18 |
+
</div><br>
|
19 |
+
|
20 |
+
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).
|
21 |
+
|
22 |
+
|
23 |
+
## Developed by
|
24 |
+
|
25 |
+
* Jiwoo Hong<sup>*</sup> (KAIST AI)
|
26 |
+
* Sayak Paul<sup>*</sup> (Hugging Face)
|
27 |
+
* Noah Lee (KAIST AI)
|
28 |
+
* Kashif Rasul (Hugging Face)
|
29 |
+
* James Thorne (KAIST AI)
|
30 |
+
* Jongheon Jeong (Korea University)
|
31 |
+
|
32 |
+
## Dataset
|
33 |
+
|
34 |
+
This model was fine-tuned from [Stable Diffusion XL](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) on the [yuvalkirstain/pickapic_v2](mhttps://huggingface.co/datasets/yuvalkirstain/pickapic_v2) dataset.
|
35 |
+
|
36 |
+
## Training Code
|
37 |
+
|
38 |
+
Refer to our code repository [here](https://github.com/mapo-t2i/mapo).
|
39 |
+
|
40 |
+
## Results
|
41 |
+
|
42 |
+
Below we report some quantitative metrics and use them to compare MaPO to existing models:
|
43 |
+
|
44 |
+
<style>
|
45 |
+
table {
|
46 |
+
width: 100%;
|
47 |
+
border-collapse: collapse;
|
48 |
+
}
|
49 |
+
th, td {
|
50 |
+
border: 1px solid #000;
|
51 |
+
padding: 8px;
|
52 |
+
text-align: center;
|
53 |
+
}
|
54 |
+
th {
|
55 |
+
background-color: #808080;
|
56 |
+
}
|
57 |
+
.ours {
|
58 |
+
font-style: italic;
|
59 |
+
}
|
60 |
+
</style>
|
61 |
+
|
62 |
+
<table>
|
63 |
+
<caption>Average score for Aesthetic, HPS v2.1, and PickScore</caption>
|
64 |
+
<thead>
|
65 |
+
<tr>
|
66 |
+
<th></th>
|
67 |
+
<th>Aesthetic</th>
|
68 |
+
<th>HPS v2.1</th>
|
69 |
+
<th>Pickscore</th>
|
70 |
+
</tr>
|
71 |
+
</thead>
|
72 |
+
<tbody>
|
73 |
+
<tr>
|
74 |
+
<td>SDXL</td>
|
75 |
+
<td>6.03</td>
|
76 |
+
<td>30.0</td>
|
77 |
+
<td>22.4</td>
|
78 |
+
</tr>
|
79 |
+
<tr>
|
80 |
+
<td>SFT<sub>Chosen</sub></td>
|
81 |
+
<td>5.95</td>
|
82 |
+
<td>29.6</td>
|
83 |
+
<td>22.0</td>
|
84 |
+
</tr>
|
85 |
+
<tr>
|
86 |
+
<td>Diffusion-DPO</td>
|
87 |
+
<td>6.03</td>
|
88 |
+
<td>31.1</td>
|
89 |
+
<td>22.7</td>
|
90 |
+
</tr>
|
91 |
+
<tr class="ours">
|
92 |
+
<td>MaPO (Ours)</td>
|
93 |
+
<td>6.17</td>
|
94 |
+
<td>31.2</td>
|
95 |
+
<td>22.5</td>
|
96 |
+
</tr>
|
97 |
+
</tbody>
|
98 |
+
</table>
|
99 |
+
|
100 |
+
|
101 |
+
We evaluated this checkpoint in the Imgsys public benchmark. MaPO was able to outperform or match 21 out of 25 state-of-the-art text-to-image diffusion models by ranking 7th on the leaderboard at the time of writing, compared to Diffusion-DPO’s 20th place, while also consuming 14.5% less wall-clock training time on adapting Pick-a-Pic v2. We appreciate the imgsys team for helping us get the human preference data.
|
102 |
+
|
103 |
+
<div align="center">
|
104 |
+
<img src="https://mapo-t2i.github.io/static/images/imgsys.png" width=750/>
|
105 |
+
</div>
|
106 |
+
|
107 |
+
|
108 |
+
## Inference
|
109 |
+
|
110 |
+
```python
|
111 |
+
from diffusers import DiffusionPipeline, AutoencoderKL, UNet2DConditionModel
|
112 |
+
import torch
|
113 |
+
|
114 |
+
sdxl_id = "stabilityai/stable-diffusion-xl-base-1.0"
|
115 |
+
vae_id = "madebyollin/sdxl-vae-fp16-fix"
|
116 |
+
unet_id = "mapo-t2i/mapo-beta"
|
117 |
+
|
118 |
+
vae = AutoencoderKL.from_pretrained(vae_id, torch_dtype=torch.float16)
|
119 |
+
unet = UNet2DConditionModel.from_pretrained(unet_id, torch_dtype=torch.float16)
|
120 |
+
pipeline = DiffusionPipeline.from_pretrained(sdxl_id, vae=vae, unet=unet, torch_dtype=torch.float16).to("cuda")
|
121 |
+
|
122 |
+
prompt = "An abstract portrait consisting of bold, flowing brushstrokes against a neutral background."
|
123 |
+
image = pipeline(prompt=prompt, num_inference_steps=30).images[0]
|
124 |
+
```
|
125 |
+
|
126 |
+
For qualitative results, please visit our [project website](https://mapo-t2i.github.io/).
|
127 |
+
|
128 |
+
## Citation
|
129 |
+
|
130 |
+
```bibtex
|
131 |
+
@misc{todo,
|
132 |
+
title={Margin-aware Preference Optimization for Aligning Diffusion Models without Reference},
|
133 |
+
author={Jiwoo Hong and Sayak Paul and Noah Lee and Kashif Rasuland James Thorne and Jongheon Jeong},
|
134 |
+
year={2024},
|
135 |
+
eprint={todo},
|
136 |
+
archivePrefix={arXiv},
|
137 |
+
primaryClass={cs.CV,cs.LG}
|
138 |
+
}
|
139 |
+
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|