ddw2AIGROUP2CQUPT
commited on
Update README.md
Browse files
README.md
CHANGED
@@ -51,6 +51,10 @@ We utilize the information of FaceCaption-15M (each image in FaceCaption-15M co
|
|
51 |
|
52 |
We present the comparisons in Table. We can make the main observations as follows: (1) In terms of the realism for generated facial images (VLM-score), our proposed Face-MakeUp significantly outperforms other models, indicating that our model can generate more realistic facial images. This is also demonstrated by the examples shown in Fig. 1. (2) Regarding attribute prediction in generated facial images (Attr c), facial images generated by FaceMakeUp contain more attributes than that of others, indicating that our model is capable of generating facial images that contain more fine-grained features. (3) In terms of similarity between generated facial images and reference (CLIP-I, DINO, FaceSim, and FID), attributed to the diversified facial feature fusion mechanism, our model achieved seven first-place and one second-place performances across two test datasets. (4) In terms of image-text similarity, our model is slightly lower than other models, mainly because the image contains not only faces but also other content. We mainly focus on optimizing the face region.
|
53 |
|
|
|
|
|
|
|
|
|
54 |
## Citation
|
55 |
|
56 |
```
|
|
|
51 |
|
52 |
We present the comparisons in Table. We can make the main observations as follows: (1) In terms of the realism for generated facial images (VLM-score), our proposed Face-MakeUp significantly outperforms other models, indicating that our model can generate more realistic facial images. This is also demonstrated by the examples shown in Fig. 1. (2) Regarding attribute prediction in generated facial images (Attr c), facial images generated by FaceMakeUp contain more attributes than that of others, indicating that our model is capable of generating facial images that contain more fine-grained features. (3) In terms of similarity between generated facial images and reference (CLIP-I, DINO, FaceSim, and FID), attributed to the diversified facial feature fusion mechanism, our model achieved seven first-place and one second-place performances across two test datasets. (4) In terms of image-text similarity, our model is slightly lower than other models, mainly because the image contains not only faces but also other content. We mainly focus on optimizing the face region.
|
53 |
|
54 |
+
## Usage
|
55 |
+
|
56 |
+
Our training and inference code have been released publicly on github.com/ddw2AIGROUP2CQUPT/Face-MakeUp(github.com)
|
57 |
+
|
58 |
## Citation
|
59 |
|
60 |
```
|