Update README.md
Browse files
README.md
CHANGED
@@ -24,4 +24,91 @@ dataset_info:
|
|
24 |
---
|
25 |
# Dataset Card for "sdxl-0.9"
|
26 |
|
27 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
---
|
25 |
# Dataset Card for "sdxl-0.9"
|
26 |
|
27 |
+
Dataset was generated using the code below:
|
28 |
+
|
29 |
+
```python
|
30 |
+
import torch
|
31 |
+
from datasets import Dataset, Features
|
32 |
+
from datasets import Image as ImageFeature
|
33 |
+
from datasets import Value, load_dataset
|
34 |
+
|
35 |
+
from diffusers import DDIMScheduler, DiffusionPipeline
|
36 |
+
import PIL
|
37 |
+
|
38 |
+
|
39 |
+
def main():
|
40 |
+
print("Loading dataset...")
|
41 |
+
parti_prompts = load_dataset("nateraw/parti-prompts", split="train")
|
42 |
+
|
43 |
+
print("Loading pipeline...")
|
44 |
+
ckpt_id = "stabilityai/stable-diffusion-xl-base-0.9"
|
45 |
+
pipe = DiffusionPipeline.from_pretrained(
|
46 |
+
ckpt_id, torch_dtype=torch.float16, use_auth_token=True
|
47 |
+
).to("cuda")
|
48 |
+
pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
|
49 |
+
pipe.set_progress_bar_config(disable=True)
|
50 |
+
|
51 |
+
seed = 0
|
52 |
+
generator = torch.Generator("cuda").manual_seed(seed)
|
53 |
+
|
54 |
+
print("Running inference...")
|
55 |
+
main_dict = {}
|
56 |
+
for i in range(len(parti_prompts)):
|
57 |
+
sample = parti_prompts[i]
|
58 |
+
prompt = sample["Prompt"]
|
59 |
+
image = pipe(
|
60 |
+
prompt,
|
61 |
+
generator=generator,
|
62 |
+
num_inference_steps=100,
|
63 |
+
guidance_scale=7.5,
|
64 |
+
).images[0]
|
65 |
+
|
66 |
+
image = image.resize((256, 256), resample=PIL.Image.Resampling.LANCZOS)
|
67 |
+
img_path = f"sd_xl_{i}.png"
|
68 |
+
image.save(img_path)
|
69 |
+
main_dict.update(
|
70 |
+
{
|
71 |
+
prompt: {
|
72 |
+
"img_path": img_path,
|
73 |
+
"Category": sample["Category"],
|
74 |
+
"Challenge": sample["Challenge"],
|
75 |
+
"Note": sample["Note"],
|
76 |
+
"model_name": ckpt_id,
|
77 |
+
"seed": seed,
|
78 |
+
}
|
79 |
+
}
|
80 |
+
)
|
81 |
+
|
82 |
+
def generation_fn():
|
83 |
+
for prompt in main_dict:
|
84 |
+
prompt_entry = main_dict[prompt]
|
85 |
+
yield {
|
86 |
+
"Prompt": prompt,
|
87 |
+
"Category": prompt_entry["Category"],
|
88 |
+
"Challenge": prompt_entry["Challenge"],
|
89 |
+
"Note": prompt_entry["Note"],
|
90 |
+
"images": {"path": prompt_entry["img_path"]},
|
91 |
+
"model_name": prompt_entry["model_name"],
|
92 |
+
"seed": prompt_entry["seed"],
|
93 |
+
}
|
94 |
+
|
95 |
+
print("Preparing HF dataset...")
|
96 |
+
ds = Dataset.from_generator(
|
97 |
+
generation_fn,
|
98 |
+
features=Features(
|
99 |
+
Prompt=Value("string"),
|
100 |
+
Category=Value("string"),
|
101 |
+
Challenge=Value("string"),
|
102 |
+
Note=Value("string"),
|
103 |
+
images=ImageFeature(),
|
104 |
+
model_name=Value("string"),
|
105 |
+
seed=Value("int64"),
|
106 |
+
),
|
107 |
+
)
|
108 |
+
ds_id = "diffusers-parti-prompts/sdxl-0.9"
|
109 |
+
ds.push_to_hub(ds_id)
|
110 |
+
|
111 |
+
|
112 |
+
if __name__ == "__main__":
|
113 |
+
main()
|
114 |
+
```
|