Spaces:
Running
on
Zero
Running
on
Zero
File size: 2,885 Bytes
4eb7414 1525ca6 4eb7414 |
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 81 82 83 84 85 86 87 88 89 90 |
import gradio as gr
import numpy as np
import torch
from PIL import Image
from gradio_image_prompter import ImagePrompter
from transformers import AutoProcessor, UdopForConditionalGeneration
import easyocr
from PIL import Image
import spaces
processor = AutoProcessor.from_pretrained("microsoft/udop-large", apply_ocr=False)
model = UdopForConditionalGeneration.from_pretrained("microsoft/udop-large")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
@spaces.GPU
def udop_box_inference(image, text_prompt, box_coordinates):
box_coordinates = [box_coordinates[0], box_coordinates[1], box_coordinates[3], box_coordinates[4]]
extracted_image = extract_box(image, box_coordinates)
extracted_image.save("cropped_image.png")
reader = easyocr.Reader(['en'])
result = reader.readtext('cropped_image.png')
texts = []
bboxs = []
for (bbox, text, prob) in result:
texts.append(text)
bboxs.append([bbox[0][0], bbox[0][1], bbox[2][0], bbox[2][1]])
height = image.size[1]
width = image.size[0]
image = image.convert("RGB")
norm_boxes = []
for box in bboxs:
norm_boxes.append(normalize_bbox(box, width, height))
encoding = processor(image, text_prompt, texts, boxes=norm_boxes, return_tensors="pt")
predicted_ids = model.generate(**encoding)
return processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
def normalize_bbox(bbox, width, height):
return [
int(1000 * (bbox[0] / width)),
int(1000 * (bbox[1] / height)),
int(1000 * (bbox[2] / width)),
int(1000 * (bbox[3] / height)),
]
def extract_box(image_path, coordinates):
x, y, x2, y2 = coordinates
cropped_image = image.crop((x, y, x2, y2))
return cropped_image
def infer_box(prompts, text_prompts):
# background (original image) layers[0] ( point prompt) composite (total image)
image = prompts["image"]
if image is None:
gr.Error("Please upload an image and draw a box before submitting")
points = prompts["points"][0]
if points is None:
gr.Error("Please draw a box before submitting.")
return udop_box_inference(image, text_prompts, points)
with gr.Blocks(title="UDOP") as demo:
gr.Markdown("# UDOP")
gr.Markdown("UDOP is a cutting-edge foundation model for a document understanding and generation.")
gr.Markdown("Try UDOP in this demo.")
with gr.Row():
with gr.Column(scale=1):
# Title
gr.Markdown("To try box prompting, simply upload and image and draw a box on it.")
with gr.Row():
with gr.Column():
im = ImagePrompter(type="pil")
text_prompt = gr.Textbox()
btn = gr.Button("Submit")
with gr.Column():
output = gr.Textbox(label="UDOP Output")
btn.click(infer_box, inputs=[im,text_prompt], outputs=[output])
demo.launch(debug=True) |