OAOA commited on
Commit
526e811
1 Parent(s): c87ab07

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +54 -5
README.md CHANGED
@@ -1,5 +1,54 @@
1
- ---
2
- license: other
3
- license_name: license
4
- license_link: LICENSE
5
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ license_name: license
4
+ license_link: LICENSE
5
+ pipeline_tag: image-to-image
6
+ tags:
7
+ - Image Super-resolution
8
+ - Diffusion Inversion
9
+ ---
10
+
11
+ # InvSR Model Card
12
+ This model card focuses on the models associated with the InvSR project, which is available [here](https://github.com/zsyOAOA/InvSR).
13
+
14
+ ## Model Details
15
+ - **Developed by:** Zongsheng Yue
16
+ - **Model type:** Arbitrary-steps Image Super-resolution via Diffusion Inversion
17
+ - **Model Description:** This is the model used in [Paper](https://arxiv.org/abs/2409.17058).
18
+ - **Resources for more information:** [GitHub Repository](https://github.com/zsyOAOA/InvSR).
19
+ - **Cite as:**
20
+
21
+ @article{yue2024invSR,
22
+ author = {Zongsheng Yue, Kang Liao, Chen Change Loy},
23
+ title = {Arbitrary-steps Image Super-resolution via Diffusion Inversion},
24
+ journal = {arxiv},
25
+ year = {2024},
26
+ }
27
+
28
+ ## Limitations and Bias
29
+
30
+ ### Limitations
31
+
32
+ - InvSR requires a tiled operation for generating a high-resolution image, which would largely increase the inference time.
33
+ - InvSR sometimes cannot keep 100% fidelity due to its generative nature.
34
+ - InvSR sometimes cannot generate perfect details under complex real-world scenarios.
35
+
36
+ ### Bias
37
+ While our model is based on a pre-trained SD-Turbo model, currently we do not observe obvious bias in generated results.
38
+
39
+ ## Training
40
+
41
+ **Training Data**
42
+ The model developer used the following dataset for training the model:
43
+
44
+ - Our model is finetuned on [LSDIR](https://data.vision.ee.ethz.ch/yawli/index.html) + 20K samples from FFHQ datasets.
45
+
46
+ **Training Procedure**
47
+ InvSR achieves the goal of image super-resolution via diffusion inversion technique on [SD-Turbo](https://huggingface.co/stabilityai/sd-turbo), detailed training pipelines can be found in our GitHub [repo](https://github.com/zsyOAOA/InvSR).
48
+
49
+ We currently provide the following checkpoints:
50
+
51
+ - [noise_predictor_sd_turbo_v5.pth](https://huggingface.co/OAOA/InvSR/blob/main/noise_predictor_sd_turbo_v5.pth): Noise estimation network trained for [SD-Turbo](https://huggingface.co/stabilityai/sd-turbo).
52
+
53
+ ## Evaluation Results
54
+ See [Paper](https://arxiv.org/abs/2409.17058) for details.