Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,11 +1,10 @@
|
|
1 |
import gradio as gr
|
2 |
from transformers import PaliGemmaForConditionalGeneration, PaliGemmaProcessor
|
3 |
-
from PIL import Image
|
4 |
import spaces
|
5 |
import torch
|
6 |
import re
|
7 |
|
8 |
-
model = PaliGemmaForConditionalGeneration.from_pretrained("gokaygokay/sd3-long-captioner").to("
|
9 |
processor = PaliGemmaProcessor.from_pretrained("gokaygokay/sd3-long-captioner")
|
10 |
|
11 |
def modify_caption(caption: str) -> str:
|
@@ -26,55 +25,46 @@ def modify_caption(caption: str) -> str:
|
|
26 |
|
27 |
def replace_fn(match):
|
28 |
return replacers[match.group(0)]
|
29 |
-
|
30 |
return re.sub(pattern, replace_fn, caption, count=1, flags=re.IGNORECASE)
|
31 |
|
|
|
32 |
def create_captions_rich(images):
|
33 |
-
# Debugging: Print out the type of 'images'
|
34 |
-
print(f"Type of 'images': {type(images)}")
|
35 |
-
if isinstance(images, tuple):
|
36 |
-
print("Received a tuple, expected a file-like object.")
|
37 |
-
# If it's a tuple, you can try accessing the first element as an example
|
38 |
-
print(f"Type of 'images[0]': {type(images[0])}")
|
39 |
-
|
40 |
captions = []
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
generation = generation[0][input_len:]
|
52 |
-
decoded = processor.decode(generation, skip_special_tokens=True)
|
53 |
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
captions.append(f"Error processing image: {e}")
|
58 |
return captions
|
59 |
|
60 |
-
|
61 |
css = """
|
62 |
#mkd {
|
63 |
height: 500px;
|
64 |
overflow: auto;
|
65 |
-
border:
|
66 |
}
|
67 |
"""
|
68 |
|
69 |
with gr.Blocks(css=css) as demo:
|
70 |
-
|
71 |
-
|
72 |
-
with gr.
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
|
|
77 |
|
78 |
-
|
79 |
|
80 |
-
demo.launch(debug=True)
|
|
|
1 |
import gradio as gr
|
2 |
from transformers import PaliGemmaForConditionalGeneration, PaliGemmaProcessor
|
|
|
3 |
import spaces
|
4 |
import torch
|
5 |
import re
|
6 |
|
7 |
+
model = PaliGemmaForConditionalGeneration.from_pretrained("gokaygokay/sd3-long-captioner").to("cuda").eval()
|
8 |
processor = PaliGemmaProcessor.from_pretrained("gokaygokay/sd3-long-captioner")
|
9 |
|
10 |
def modify_caption(caption: str) -> str:
|
|
|
25 |
|
26 |
def replace_fn(match):
|
27 |
return replacers[match.group(0)]
|
28 |
+
|
29 |
return re.sub(pattern, replace_fn, caption, count=1, flags=re.IGNORECASE)
|
30 |
|
31 |
+
@spaces.GPU
|
32 |
def create_captions_rich(images):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
captions = []
|
34 |
+
prompt = "caption en"
|
35 |
+
|
36 |
+
for image in images:
|
37 |
+
model_inputs = processor(text=prompt, images=image, return_tensors="pt").to("cuda")
|
38 |
+
input_len = model_inputs["input_ids"].shape[-1]
|
39 |
+
|
40 |
+
with torch.inference_mode():
|
41 |
+
generation = model.generate(**model_inputs, max_new_tokens=256, do_sample=False)
|
42 |
+
generation = generation[0][input_len:]
|
43 |
+
decoded = processor.decode(generation, skip_special_tokens=True)
|
|
|
|
|
44 |
|
45 |
+
modified_caption = modify_caption(decoded)
|
46 |
+
captions.append(modified_caption)
|
47 |
+
|
|
|
48 |
return captions
|
49 |
|
|
|
50 |
css = """
|
51 |
#mkd {
|
52 |
height: 500px;
|
53 |
overflow: auto;
|
54 |
+
border: 16px solid #ccc;
|
55 |
}
|
56 |
"""
|
57 |
|
58 |
with gr.Blocks(css=css) as demo:
|
59 |
+
gr.HTML("<h1><center>Fine-tuned PaliGemma for SD3 Image Guided Prompt Generation.<center><h1>")
|
60 |
+
|
61 |
+
with gr.Tab(label="Image to Prompt for SD3."):
|
62 |
+
with gr.Row():
|
63 |
+
with gr.Column():
|
64 |
+
input_imgs = gr.Image(label="Input Images", type="pil", tool="editor", interactive=True, multiple=True)
|
65 |
+
submit_btn = gr.Button(value="Start")
|
66 |
+
outputs = gr.Text(label="Prompts", interactive=False)
|
67 |
|
68 |
+
submit_btn.click(create_captions_rich, [input_imgs], [outputs])
|
69 |
|
70 |
+
demo.launch(debug=True)
|