Spaces:
Running
Running
File size: 2,888 Bytes
b502a48 8308143 b502a48 13bac9e b502a48 80b7ee0 b502a48 13bac9e 80b7ee0 b502a48 13bac9e b502a48 3b8df03 b502a48 ebe70b1 b502a48 ebe70b1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 |
import gradio as gr
from transformers import PaliGemmaForConditionalGeneration, PaliGemmaProcessor
from PIL import Image
import spaces
import torch
import re
model = PaliGemmaForConditionalGeneration.from_pretrained("gokaygokay/sd3-long-captioner").to("cpu").eval()
processor = PaliGemmaProcessor.from_pretrained("gokaygokay/sd3-long-captioner")
def modify_caption(caption: str) -> str:
"""
Removes specific prefixes from captions.
Args:
caption (str): A string containing a caption.
Returns:
str: The caption with the prefix removed if it was present.
"""
prefix_substrings = [
('captured from ', ''),
('captured at ', '')
]
pattern = '|'.join([re.escape(opening) for opening, _ in prefix_substrings])
replacers = {opening: replacer for opening, replacer in prefix_substrings}
def replace_fn(match):
return replacers[match.group(0)]
return re.sub(pattern, replace_fn, caption, count=1, flags=re.IGNORECASE)
def create_captions_rich(images):
# Debugging: Print out the type of 'images'
print(f"Type of 'images': {type(images)}")
if isinstance(images, tuple):
print("Received a tuple, expected a file-like object.")
# If it's a tuple, you can try accessing the first element as an example
print(f"Type of 'images[0]': {type(images[0])}")
captions = []
for image_path in images:
try:
# If 'images' is a tuple, you might need to modify this part to extract the image file correctly
image = Image.open(image_path).convert("RGB")
prompt = "caption en"
model_inputs = processor(text=prompt, images=image, return_tensors="pt").to("cpu")
input_len = model_inputs["input_ids"].shape[-1]
with torch.inference_mode():
generation = model.generate(**model_inputs, max_new_tokens=256, do_sample=False)
generation = generation[0][input_len:]
decoded = processor.decode(generation, skip_special_tokens=True)
modified_caption = modify_caption(decoded)
captions.append(modified_caption)
except Exception as e:
captions.append(f"Error processing image: {e}")
return captions
css = """
#mkd {
height: 500px;
overflow: auto;
border: 8px solid #ccc;
}
"""
with gr.Blocks(css=css) as demo:
gr.HTML("<h1><center>Finetuned PaliGemma for SD3 prompt generation.<center><h1>")
with gr.Tab(label="Image to Prompt for SD3"):
with gr.Row():
with gr.Column():
input_img = gr.Gallery(label="Input Images", type="pil", interactive=True)
submit_btn = gr.Button(value="Start")
output = gr.Textbox(label="Prompt", lines=10, interactive=True)
submit_btn.click(create_captions_rich, [input_img], [output])
demo.launch(debug=True) |