EpsilonGreedy
commited on
Commit
•
9b823b1
1
Parent(s):
af1cf50
Update README.md
Browse files
README.md
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
# StableChair Diffusion
|
2 |
-
Code: https://github.com/AmosDinh/StableDiffusion_Chairs <br
|
3 |
|
4 |
The Stable Diffusion 1.5 model is finetuned on a scraped dataset which consists of chairs and their text descriptions (n=6000, german text descriptions). Finetuning is successful in the sense that the finetuned model, in contrast to the base model, creates chairs easily when supplied with (german) text it has been finetuned on. <br><br>
|
5 |
We filter for images containing only one chair using llava-phi3 model. Although the llava-phi3 model is only 3 billion parameters it performs better than the llava1.6/llava-next model (7b) on this simple counting task. Paligemma (Google) is too slow.
|
|
|
1 |
# StableChair Diffusion
|
2 |
+
Code: https://github.com/AmosDinh/StableDiffusion_Chairs <br>
|
3 |
|
4 |
The Stable Diffusion 1.5 model is finetuned on a scraped dataset which consists of chairs and their text descriptions (n=6000, german text descriptions). Finetuning is successful in the sense that the finetuned model, in contrast to the base model, creates chairs easily when supplied with (german) text it has been finetuned on. <br><br>
|
5 |
We filter for images containing only one chair using llava-phi3 model. Although the llava-phi3 model is only 3 billion parameters it performs better than the llava1.6/llava-next model (7b) on this simple counting task. Paligemma (Google) is too slow.
|