Update app.py
Browse files
app.py
CHANGED
@@ -2,135 +2,206 @@ import gradio as gr
|
|
2 |
import cv2
|
3 |
import numpy as np
|
4 |
import os
|
5 |
-
import
|
6 |
-
|
|
|
|
|
|
|
|
|
7 |
|
8 |
-
#
|
9 |
-
|
10 |
|
11 |
-
#
|
12 |
-
|
13 |
-
|
14 |
|
15 |
-
|
16 |
-
try:
|
17 |
-
model = YOLO(model_path)
|
18 |
-
except Exception as e:
|
19 |
-
raise RuntimeError(f"Failed to load Latex2Layout model: {e}")
|
20 |
-
|
21 |
-
def detect_and_visualize(image):
|
22 |
"""
|
23 |
-
|
24 |
-
|
25 |
Args:
|
26 |
-
|
27 |
-
|
28 |
Returns:
|
29 |
-
|
30 |
-
yolo_annotations: Annotations in YOLO format as a string.
|
31 |
"""
|
32 |
-
#
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
try:
|
38 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
except Exception as e:
|
40 |
-
|
41 |
-
|
42 |
-
# Extract results from the first frame
|
43 |
-
result = results[0]
|
44 |
-
annotated_image = image.copy()
|
45 |
-
yolo_annotations = []
|
46 |
-
|
47 |
-
# Get image dimensions
|
48 |
-
img_height, img_width = image.shape[:2]
|
49 |
-
|
50 |
-
# Process each detected object
|
51 |
-
for box in result.boxes:
|
52 |
-
# Extract bounding box coordinates
|
53 |
-
x1, y1, x2, y2 = box.xyxy[0].cpu().numpy()
|
54 |
-
x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2)
|
55 |
-
|
56 |
-
# Get confidence and class details
|
57 |
-
conf = float(box.conf[0])
|
58 |
-
cls_id = int(box.cls[0])
|
59 |
-
cls_name = result.names[cls_id]
|
60 |
-
|
61 |
-
# Assign a random color to the class
|
62 |
-
color = tuple(np.random.randint(0, 255, 3).tolist())
|
63 |
|
64 |
-
|
65 |
-
cv2.rectangle(annotated_image, (x1, y1), (x2, y2), color, 2)
|
66 |
-
|
67 |
-
# Create and draw label with confidence
|
68 |
-
label = f"{cls_name} {conf:.2f}"
|
69 |
-
(label_width, label_height), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)
|
70 |
-
cv2.rectangle(annotated_image, (x1, y1 - label_height - 5), (x1 + label_width, y1), color, -1)
|
71 |
-
cv2.putText(annotated_image, label, (x1, y1 - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)
|
72 |
-
|
73 |
-
# Convert bounding box to YOLO format (normalized coordinates)
|
74 |
-
x_center = (x1 + x2) / (2 * img_width)
|
75 |
-
y_center = (y1 + y2) / (2 * img_height)
|
76 |
-
width = (x2 - x1) / img_width
|
77 |
-
height = (y2 - y1) / img_height
|
78 |
-
yolo_annotations.append(f"{cls_id} {x_center:.6f} {y_center:.6f} {width:.6f} {height:.6f}")
|
79 |
-
|
80 |
-
# Combine annotations into a single string
|
81 |
-
yolo_annotations_str = "\n".join(yolo_annotations) if yolo_annotations else "No objects detected."
|
82 |
-
return annotated_image, yolo_annotations_str
|
83 |
-
|
84 |
-
def save_yolo_annotations(yolo_annotations_str):
|
85 |
"""
|
86 |
-
|
87 |
-
|
88 |
Args:
|
89 |
-
|
90 |
-
|
|
|
|
|
|
|
91 |
Returns:
|
92 |
-
|
93 |
"""
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
|
|
|
|
|
|
|
|
99 |
try:
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
except Exception as e:
|
106 |
-
|
107 |
-
|
108 |
-
# Build the Gradio interface
|
109 |
-
with gr.Blocks(title="Latex2Layout Object Detection Visualization") as demo:
|
110 |
-
gr.Markdown("# Latex2Layout Object Detection Visualization")
|
111 |
-
gr.Markdown("Upload an image to detect objects using the Latex2Layout model. View the results with bounding boxes and download annotations in YOLO format.")
|
112 |
|
|
|
|
|
|
|
|
|
|
|
113 |
with gr.Row():
|
114 |
-
with gr.Column():
|
115 |
input_image = gr.Image(label="Upload Image", type="numpy")
|
116 |
-
detect_btn = gr.Button("
|
|
|
117 |
|
118 |
-
with gr.Column():
|
119 |
output_image = gr.Image(label="Detection Results")
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
detect_btn.click(
|
126 |
-
fn=
|
127 |
inputs=[input_image],
|
128 |
-
outputs=[output_image,
|
129 |
)
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
|
|
134 |
)
|
135 |
|
136 |
# Launch the application
|
|
|
2 |
import cv2
|
3 |
import numpy as np
|
4 |
import os
|
5 |
+
import requests
|
6 |
+
import json
|
7 |
+
from PIL import Image
|
8 |
+
import io
|
9 |
+
import base64
|
10 |
+
from openai import OpenAI
|
11 |
|
12 |
+
# API endpoints
|
13 |
+
YOLO_API_ENDPOINT = "https://api.example.com/yolo" # Replace with actual YOLO API endpoint
|
14 |
|
15 |
+
# Qwen API configuration
|
16 |
+
QWEN_BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1"
|
17 |
+
QWEN_MODEL_ID = "qwen2.5-vl-3b-instruct"
|
18 |
|
19 |
+
def encode_image(image_array):
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
"""
|
21 |
+
Encode numpy array image to base64 string.
|
22 |
+
|
23 |
Args:
|
24 |
+
image_array: numpy array of the image
|
25 |
+
|
26 |
Returns:
|
27 |
+
base64 encoded string of the image
|
|
|
28 |
"""
|
29 |
+
# Convert numpy array to PIL Image
|
30 |
+
pil_image = Image.fromarray(image_array)
|
31 |
+
|
32 |
+
# Convert PIL Image to bytes
|
33 |
+
img_byte_arr = io.BytesIO()
|
34 |
+
pil_image.save(img_byte_arr, format='PNG')
|
35 |
+
img_byte_arr = img_byte_arr.getvalue()
|
36 |
+
|
37 |
+
# Encode to base64
|
38 |
+
return base64.b64encode(img_byte_arr).decode("utf-8")
|
39 |
+
|
40 |
+
def detect_layout(image):
|
41 |
+
"""
|
42 |
+
Perform layout detection on the uploaded image using YOLO API.
|
43 |
+
|
44 |
+
Args:
|
45 |
+
image: The uploaded image as a numpy array
|
46 |
+
|
47 |
+
Returns:
|
48 |
+
annotated_image: Image with detection boxes
|
49 |
+
layout_info: Layout detection results
|
50 |
+
"""
|
51 |
+
if image is None:
|
52 |
+
return None, "Error: No image uploaded."
|
53 |
+
|
54 |
+
# Convert numpy array to PIL Image
|
55 |
+
pil_image = Image.fromarray(image)
|
56 |
+
|
57 |
+
# Convert PIL Image to bytes for API request
|
58 |
+
img_byte_arr = io.BytesIO()
|
59 |
+
pil_image.save(img_byte_arr, format='PNG')
|
60 |
+
img_byte_arr = img_byte_arr.getvalue()
|
61 |
+
|
62 |
+
# Prepare API request
|
63 |
+
files = {'image': ('image.png', img_byte_arr, 'image/png')}
|
64 |
+
|
65 |
try:
|
66 |
+
# Call YOLO API
|
67 |
+
response = requests.post(YOLO_API_ENDPOINT, files=files)
|
68 |
+
response.raise_for_status()
|
69 |
+
detection_results = response.json()
|
70 |
+
|
71 |
+
# Create a copy of the image for visualization
|
72 |
+
annotated_image = image.copy()
|
73 |
+
|
74 |
+
# Draw detection results
|
75 |
+
for detection in detection_results:
|
76 |
+
x1, y1, x2, y2 = detection['bbox']
|
77 |
+
cls_name = detection['class']
|
78 |
+
conf = detection['confidence']
|
79 |
+
|
80 |
+
# Generate a color for each class
|
81 |
+
color = tuple(np.random.randint(0, 255, 3).tolist())
|
82 |
+
|
83 |
+
# Draw bounding box and label
|
84 |
+
cv2.rectangle(annotated_image, (int(x1), int(y1)), (int(x2), int(y2)), color, 2)
|
85 |
+
label = f'{cls_name} {conf:.2f}'
|
86 |
+
(label_width, label_height), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)
|
87 |
+
cv2.rectangle(annotated_image, (int(x1), int(y1)-label_height-5), (int(x1)+label_width, int(y1)), color, -1)
|
88 |
+
cv2.putText(annotated_image, label, (int(x1), int(y1)-5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)
|
89 |
+
|
90 |
+
# Format layout information for Qwen
|
91 |
+
layout_info = json.dumps(detection_results, indent=2)
|
92 |
+
|
93 |
+
return annotated_image, layout_info
|
94 |
+
|
95 |
except Exception as e:
|
96 |
+
return None, f"Error during layout detection: {str(e)}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
97 |
|
98 |
+
def qa_about_layout(image, question, layout_info, api_key):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
"""
|
100 |
+
Answer questions about the layout using Qwen2.5-VL API.
|
101 |
+
|
102 |
Args:
|
103 |
+
image: The uploaded image
|
104 |
+
question: User's question about the layout
|
105 |
+
layout_info: Layout detection results from YOLO
|
106 |
+
api_key: User's Qwen API key
|
107 |
+
|
108 |
Returns:
|
109 |
+
answer: Qwen's answer to the question
|
110 |
"""
|
111 |
+
if image is None or not question:
|
112 |
+
return "Please upload an image and ask a question."
|
113 |
+
|
114 |
+
if not layout_info:
|
115 |
+
return "No layout information available. Please detect layout first."
|
116 |
+
|
117 |
+
if not api_key:
|
118 |
+
return "Please enter your Qwen API key."
|
119 |
+
|
120 |
try:
|
121 |
+
# Encode image to base64
|
122 |
+
base64_image = encode_image(image)
|
123 |
+
|
124 |
+
# Initialize OpenAI client for Qwen API
|
125 |
+
client = OpenAI(
|
126 |
+
api_key=api_key,
|
127 |
+
base_url=QWEN_BASE_URL,
|
128 |
+
)
|
129 |
+
|
130 |
+
# Prepare system prompt with layout information
|
131 |
+
system_prompt = f"""You are a helpful assistant specialized in analyzing document layouts.
|
132 |
+
The following layout information has been detected in the image:
|
133 |
+
{layout_info}
|
134 |
+
|
135 |
+
Please answer questions about the layout based on this information and the image."""
|
136 |
+
|
137 |
+
# Prepare messages for API call
|
138 |
+
messages = [
|
139 |
+
{
|
140 |
+
"role": "system",
|
141 |
+
"content": [{"type": "text", "text": system_prompt}]
|
142 |
+
},
|
143 |
+
{
|
144 |
+
"role": "user",
|
145 |
+
"content": [
|
146 |
+
{
|
147 |
+
"type": "image_url",
|
148 |
+
"image_url": {"url": f"data:image/jpeg;base64,{base64_image}"},
|
149 |
+
},
|
150 |
+
{"type": "text", "text": question},
|
151 |
+
],
|
152 |
+
}
|
153 |
+
]
|
154 |
+
|
155 |
+
# Call Qwen API
|
156 |
+
completion = client.chat.completions.create(
|
157 |
+
model=QWEN_MODEL_ID,
|
158 |
+
messages=messages,
|
159 |
+
)
|
160 |
+
|
161 |
+
return completion.choices[0].message.content
|
162 |
+
|
163 |
except Exception as e:
|
164 |
+
return f"Error during QA: {str(e)}"
|
|
|
|
|
|
|
|
|
|
|
165 |
|
166 |
+
# Create Gradio interface
|
167 |
+
with gr.Blocks(title="Latex2Layout QA System") as demo:
|
168 |
+
gr.Markdown("# Latex2Layout QA System")
|
169 |
+
gr.Markdown("Upload an image, detect layout elements, and ask questions about the layout.")
|
170 |
+
|
171 |
with gr.Row():
|
172 |
+
with gr.Column(scale=1):
|
173 |
input_image = gr.Image(label="Upload Image", type="numpy")
|
174 |
+
detect_btn = gr.Button("Detect Layout")
|
175 |
+
gr.Markdown("**Tip**: Upload a clear image for optimal detection results.")
|
176 |
|
177 |
+
with gr.Column(scale=1):
|
178 |
output_image = gr.Image(label="Detection Results")
|
179 |
+
layout_info = gr.Textbox(label="Layout Information", lines=10)
|
180 |
+
|
181 |
+
with gr.Row():
|
182 |
+
with gr.Column(scale=1):
|
183 |
+
api_key_input = gr.Textbox(
|
184 |
+
label="Qwen API Key",
|
185 |
+
placeholder="Enter your Qwen API key here",
|
186 |
+
type="password"
|
187 |
+
)
|
188 |
+
question_input = gr.Textbox(label="Ask a question about the layout")
|
189 |
+
qa_btn = gr.Button("Ask Question")
|
190 |
+
|
191 |
+
with gr.Column(scale=1):
|
192 |
+
answer_output = gr.Textbox(label="Answer", lines=5)
|
193 |
+
|
194 |
+
# Event handlers
|
195 |
detect_btn.click(
|
196 |
+
fn=detect_layout,
|
197 |
inputs=[input_image],
|
198 |
+
outputs=[output_image, layout_info]
|
199 |
)
|
200 |
+
|
201 |
+
qa_btn.click(
|
202 |
+
fn=qa_about_layout,
|
203 |
+
inputs=[input_image, question_input, layout_info, api_key_input],
|
204 |
+
outputs=[answer_output]
|
205 |
)
|
206 |
|
207 |
# Launch the application
|