Spaces:
Sleeping
Sleeping
KoonJamesZ
commited on
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,193 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import TableTransformerForObjectDetection
|
4 |
+
import matplotlib.pyplot as plt
|
5 |
+
from transformers import DetrFeatureExtractor
|
6 |
+
import pandas as pd
|
7 |
+
import uuid
|
8 |
+
from surya.ocr import run_ocr
|
9 |
+
# from surya.model.detection.segformer import load_model as load_det_model, load_processor as load_det_processor
|
10 |
+
from surya.model.detection.model import load_model as load_det_model, load_processor as load_det_processor
|
11 |
+
from surya.model.recognition.model import load_model as load_rec_model
|
12 |
+
from surya.model.recognition.processor import load_processor as load_rec_processor
|
13 |
+
from PIL import ImageDraw, Image
|
14 |
+
import os
|
15 |
+
from pdf2image import convert_from_path
|
16 |
+
import tempfile
|
17 |
+
from ultralyticsplus import YOLO, render_result
|
18 |
+
import cv2
|
19 |
+
import numpy as np
|
20 |
+
from fpdf import FPDF
|
21 |
+
|
22 |
+
def convert_pdf_images(pdf_path):
|
23 |
+
# Convert PDF to images
|
24 |
+
images = convert_from_path(pdf_path)
|
25 |
+
|
26 |
+
# Save each page as a temporary image and collect file paths
|
27 |
+
temp_file_paths = []
|
28 |
+
for i, page in enumerate(images):
|
29 |
+
# Create a temporary file with a unique name
|
30 |
+
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".png")
|
31 |
+
page.save(temp_file.name, 'PNG') # Save the image to the temporary file
|
32 |
+
temp_file_paths.append(temp_file.name) # Add file path to the list
|
33 |
+
|
34 |
+
return temp_file_paths[0] # Return the list of temporary file paths
|
35 |
+
|
36 |
+
|
37 |
+
# Load model
|
38 |
+
model_yolo = YOLO('keremberke/yolov8m-table-extraction')
|
39 |
+
|
40 |
+
# Set model parameters
|
41 |
+
model_yolo.overrides['conf'] = 0.25 # NMS confidence threshold
|
42 |
+
model_yolo.overrides['iou'] = 0.45 # NMS IoU threshold
|
43 |
+
model_yolo.overrides['agnostic_nms'] = False # NMS class-agnostic
|
44 |
+
model_yolo.overrides['max_det'] = 1000 # maximum number of detections per image
|
45 |
+
|
46 |
+
# new v1.1 checkpoints require no timm anymore
|
47 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
48 |
+
langs = ["en","th"] # Replace with your languages - optional but recommended
|
49 |
+
det_processor, det_model = load_det_processor(), load_det_model()
|
50 |
+
rec_model, rec_processor = load_rec_model(), load_rec_processor()
|
51 |
+
|
52 |
+
|
53 |
+
feature_extractor = DetrFeatureExtractor()
|
54 |
+
|
55 |
+
model = TableTransformerForObjectDetection.from_pretrained("microsoft/table-transformer-structure-recognition-v1.1-all")
|
56 |
+
|
57 |
+
def crop_table(filename):
|
58 |
+
# Set image
|
59 |
+
image_path = filename
|
60 |
+
image = Image.open(image_path)
|
61 |
+
image_np = np.array(image)
|
62 |
+
|
63 |
+
# Perform inference
|
64 |
+
results = model_yolo.predict(image_path)
|
65 |
+
|
66 |
+
# Extract the first bounding box (assuming there's only one table)
|
67 |
+
bbox = results[0].boxes[0]
|
68 |
+
x1, y1, x2, y2 = map(int, bbox.xyxy[0]) # Get the bounding box coordinates
|
69 |
+
|
70 |
+
# Crop the image using the bounding box coordinates
|
71 |
+
cropped_image = image_np[y1:y2, x1:x2]
|
72 |
+
|
73 |
+
# Convert the cropped image to RGB (if it's not already in RGB)
|
74 |
+
cropped_image_rgb = cv2.cvtColor(cropped_image, cv2.COLOR_BGR2RGB)
|
75 |
+
|
76 |
+
# Save the cropped image as a PDF
|
77 |
+
cropped_image_pil = Image.fromarray(cropped_image_rgb)
|
78 |
+
# Save the cropped image to a temporary file
|
79 |
+
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".png")
|
80 |
+
cropped_image_pil.save(temp_file.name)
|
81 |
+
|
82 |
+
return temp_file.name
|
83 |
+
|
84 |
+
def extract_table(image_path):
|
85 |
+
image = Image.open(image_path)
|
86 |
+
predictions = run_ocr([image], [langs], det_model, det_processor, rec_model, rec_processor)
|
87 |
+
objs = []
|
88 |
+
for t in predictions[0].text_lines:
|
89 |
+
objs.append([t.polygon,t.confidence,t.text,t.bbox])
|
90 |
+
# Sort objects by their y-coordinate to facilitate row separation
|
91 |
+
objs = sorted(objs, key=lambda x: x[3][1])
|
92 |
+
|
93 |
+
# Initialize lists to store rows and column boundaries
|
94 |
+
rows = []
|
95 |
+
row_threshold = 5 # Adjust as needed to separate rows based on y-coordinates
|
96 |
+
column_boundaries = []
|
97 |
+
|
98 |
+
# First pass to determine approximate column boundaries based on x-coordinates
|
99 |
+
for obj in objs:
|
100 |
+
x_min = obj[3][0] # x-coordinate of the left side of the bounding box
|
101 |
+
if not any(abs(x - x_min) < 10 for x in column_boundaries):
|
102 |
+
column_boundaries.append(x_min)
|
103 |
+
|
104 |
+
# Sort column boundaries to ensure proper left-to-right order
|
105 |
+
column_boundaries.sort()
|
106 |
+
|
107 |
+
# Second pass to organize text by rows and columns
|
108 |
+
current_row = []
|
109 |
+
previous_y = None
|
110 |
+
|
111 |
+
for obj in objs:
|
112 |
+
bbox = obj[3]
|
113 |
+
text = obj[2]
|
114 |
+
|
115 |
+
# Check if the current item belongs to a new row based on y-coordinate
|
116 |
+
if previous_y is None or abs(bbox[1] - previous_y) > row_threshold:
|
117 |
+
# Add the completed row to the list if it's not empty
|
118 |
+
if current_row:
|
119 |
+
rows.append(current_row)
|
120 |
+
current_row = [''] * len(column_boundaries) # Initialize new row with placeholders
|
121 |
+
|
122 |
+
# Find the appropriate column for the current text based on x-coordinate
|
123 |
+
for col_index, x_bound in enumerate(column_boundaries):
|
124 |
+
if abs(bbox[0] - x_bound) < 10: # Adjust threshold as necessary
|
125 |
+
current_row[col_index] = text
|
126 |
+
|
127 |
+
break
|
128 |
+
|
129 |
+
previous_y = bbox[1]
|
130 |
+
|
131 |
+
# Add the last row if it's not empty
|
132 |
+
if current_row:
|
133 |
+
rows.append(current_row)
|
134 |
+
|
135 |
+
# Create DataFrame from rows
|
136 |
+
df = pd.DataFrame(rows)
|
137 |
+
df.columns = df.iloc[0]
|
138 |
+
df = df.iloc[1:]
|
139 |
+
# Save DataFrame to an CSV file
|
140 |
+
csv_path = f'{uuid.uuid4()}.csv'
|
141 |
+
|
142 |
+
df.to_csv(csv_path,index=False)
|
143 |
+
|
144 |
+
# Save table_with_bbox_path
|
145 |
+
table_with_bbox_path = f"{uuid.uuid4()}.png"
|
146 |
+
|
147 |
+
for obj in objs:
|
148 |
+
# draw bbox on image
|
149 |
+
draw = ImageDraw.Draw(image)
|
150 |
+
draw.rectangle(obj[3], outline='red', width=1)
|
151 |
+
image.save(table_with_bbox_path)
|
152 |
+
|
153 |
+
return csv_path,table_with_bbox_path
|
154 |
+
|
155 |
+
|
156 |
+
|
157 |
+
# Function to process the uploaded file
|
158 |
+
def process_file(uploaded_file):
|
159 |
+
images_table = convert_pdf_images(uploaded_file)
|
160 |
+
croped_table = crop_table(images_table)
|
161 |
+
|
162 |
+
filepath,bbox_table= extract_table(croped_table)
|
163 |
+
|
164 |
+
os.remove(images_table)
|
165 |
+
os.remove(croped_table)
|
166 |
+
return filepath, bbox_table # Return the file path for download
|
167 |
+
|
168 |
+
# Function to clear the inputs and outputs
|
169 |
+
def clear_inputs():
|
170 |
+
return None, None, None # Clear both input and output
|
171 |
+
|
172 |
+
# Define the Gradio interface
|
173 |
+
with gr.Blocks() as demo:
|
174 |
+
gr.Markdown("## Upload a PDF, Process it, and Download the Processed File")
|
175 |
+
|
176 |
+
with gr.Row():
|
177 |
+
upload = gr.File(label="Upload PDF", type="filepath", file_types=[".pdf"])
|
178 |
+
download = gr.File(label="Download Processed PDF")
|
179 |
+
with gr.Row():
|
180 |
+
process_button = gr.Button("Process")
|
181 |
+
clear_button = gr.Button("Clear") # Custom clear button
|
182 |
+
image_display = gr.Image(label="Processed Image")
|
183 |
+
|
184 |
+
# Trigger the file processing with the button click
|
185 |
+
process_button.click(process_file, inputs=upload, outputs=[download, image_display])
|
186 |
+
|
187 |
+
# Trigger clearing inputs and outputs
|
188 |
+
clear_button.click(clear_inputs, inputs=None, outputs=[upload, download, image_display])
|
189 |
+
|
190 |
+
# Launch the interface
|
191 |
+
demo.launch()
|
192 |
+
|
193 |
+
# print(process_file("/content/ขอ ตารางกริยาช่องที่ 1 ในภาษาไทย (กริยาคำกริยา) ซ... - ขอ ตารางกริยาช่องที่ 1 ในภาษาไทย (กริย.pdf"))
|