Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,214 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python
|
2 |
+
# coding: utf-8
|
3 |
+
|
4 |
+
# In[8]:
|
5 |
+
|
6 |
+
|
7 |
+
import tensorflow as tf
|
8 |
+
from tensorflow.keras.applications.resnet50 import preprocess_input, decode_predictions
|
9 |
+
from tensorflow.keras.preprocessing import image
|
10 |
+
from ultralytics import YOLO
|
11 |
+
import numpy as np
|
12 |
+
from PIL import Image, ImageDraw, ImageFont
|
13 |
+
import gradio as gr
|
14 |
+
from huggingface_hub import snapshot_download
|
15 |
+
import os
|
16 |
+
from torchvision import transforms
|
17 |
+
|
18 |
+
# Define the class labels
|
19 |
+
classes = {0: "Defective", 1: "Good"}
|
20 |
+
|
21 |
+
model_path = "ResNet50_Classification.h5" # Trained RestNet50 model
|
22 |
+
|
23 |
+
best_yolo_model = "best.pt" # Trained YOLOv8 detection model
|
24 |
+
|
25 |
+
classification_model = tf.keras.models.load_model('ResNet50_Classification.h5')
|
26 |
+
|
27 |
+
detection_model = YOLO(best_yolo_model, task='detect')
|
28 |
+
|
29 |
+
|
30 |
+
# Define the image preprocessing function
|
31 |
+
def preprocess_imageXX(image):
|
32 |
+
image = image.resize((224, 224)) # Resize to the input size of ResNet50
|
33 |
+
image = np.array(image) # Convert to numpy array
|
34 |
+
image = preprocess_input(image) # Preprocess for ResNet50
|
35 |
+
image = np.expand_dims(image, axis=0) # Add batch dimension
|
36 |
+
return image
|
37 |
+
|
38 |
+
def preprocess_image(pilimg):
|
39 |
+
img = pilimg.resize((224, 224)) # Resize to the input size of ResNet50
|
40 |
+
img_array = image.img_to_array(img)
|
41 |
+
img_array = np.expand_dims(img_array, axis=0) # Add batch dimension
|
42 |
+
return img_array
|
43 |
+
|
44 |
+
def classify_image(pilimg):
|
45 |
+
img_array = preprocess_image(pilimg) # Preprocess the input image
|
46 |
+
classify_result = classification_model.predict(img_array)[0][0] # Get prediction probability
|
47 |
+
print(">>> Result : ", classify_result)
|
48 |
+
|
49 |
+
predicted_class = "Good" if classify_result >= 0.5 else "Defective"
|
50 |
+
print(">>> predicted_class : ", predicted_class)
|
51 |
+
|
52 |
+
return predicted_class
|
53 |
+
|
54 |
+
def detect_defect(img):
|
55 |
+
detection_result = detection_model.predict(img, conf=0.4, iou=0.5)
|
56 |
+
|
57 |
+
return detection_result
|
58 |
+
|
59 |
+
|
60 |
+
def process_image(pilimg):
|
61 |
+
# Perform classification first, then perform detection if Defective
|
62 |
+
classification = classify_image(pilimg)
|
63 |
+
|
64 |
+
if classification == "Good":
|
65 |
+
out_pilimg = pilimg.convert("RGB")
|
66 |
+
draw = ImageDraw.Draw(out_pilimg)
|
67 |
+
font = ImageFont.truetype("arialbd.ttf", 30) # Use arial.ttf for bold font
|
68 |
+
|
69 |
+
draw.text((250, 10), "Good", fill="green", font=font)
|
70 |
+
gr.Info("No defect detected,GOOD!", duration=3)
|
71 |
+
|
72 |
+
else: # Defective
|
73 |
+
detection_result = detect_defect(pilimg)
|
74 |
+
img_bgr = detection_result[0].plot()
|
75 |
+
out_pilimg = Image.fromarray(img_bgr[..., ::-1]) # RGB-order PIL image
|
76 |
+
|
77 |
+
draw = ImageDraw.Draw(out_pilimg)
|
78 |
+
font = ImageFont.truetype("arialbd.ttf", 30) # Use arial.ttf for bold font
|
79 |
+
|
80 |
+
draw.text((250, 10), "Defective", fill="red", font=font)
|
81 |
+
|
82 |
+
detections = detection_result[0].boxes.data # Get detections
|
83 |
+
if len(detections) > 0:
|
84 |
+
gr.Warning("Defect detected, BAD!", duration=3)
|
85 |
+
else:
|
86 |
+
gr.Warning("Classified as Defective but defect cannot be detected, ERROR!")
|
87 |
+
|
88 |
+
return out_pilimg
|
89 |
+
|
90 |
+
title = "Detect the status of the cap, DEFECTIVE or GOOD"
|
91 |
+
interface = gr.Interface(
|
92 |
+
fn=process_image,
|
93 |
+
inputs=gr.Image(type="pil", label="Input Image"),
|
94 |
+
outputs=gr.Image(type="pil", label="Classification result"),
|
95 |
+
title=title,
|
96 |
+
)
|
97 |
+
|
98 |
+
# Launch the interface
|
99 |
+
interface.launch(share=True)
|
100 |
+
|
101 |
+
|
102 |
+
# In[ ]:
|
103 |
+
|
104 |
+
|
105 |
+
|
106 |
+
|
107 |
+
|
108 |
+
# In[ ]:
|
109 |
+
|
110 |
+
|
111 |
+
|
112 |
+
|
113 |
+
|
114 |
+
# In[ ]:
|
115 |
+
|
116 |
+
|
117 |
+
|
118 |
+
|
119 |
+
|
120 |
+
# In[37]:
|
121 |
+
|
122 |
+
|
123 |
+
from ultralytics import YOLO
|
124 |
+
from PIL import Image, ImageDraw, ImageFont
|
125 |
+
import gradio as gr
|
126 |
+
from huggingface_hub import snapshot_download
|
127 |
+
import os
|
128 |
+
from torchvision import transforms
|
129 |
+
|
130 |
+
classes = {0: "Defective", 1: "Good"}
|
131 |
+
|
132 |
+
model_path = "best_int8_openvino_model"
|
133 |
+
|
134 |
+
def load_model_local():
|
135 |
+
detection_model = YOLO(model_path, task='classify') # Load the model
|
136 |
+
return detection_model
|
137 |
+
|
138 |
+
def load_model(repo_id):
|
139 |
+
download_dir = snapshot_download(repo_id)
|
140 |
+
print(download_dir)
|
141 |
+
path = os.path.join(download_dir, "best_int8_openvino_model")
|
142 |
+
print(path)
|
143 |
+
detection_model = YOLO(path, task='classify')
|
144 |
+
return detection_model
|
145 |
+
|
146 |
+
def predict(pilimg):
|
147 |
+
source = pilimg
|
148 |
+
|
149 |
+
# Call the model to transform image size
|
150 |
+
transform = transforms.Compose([
|
151 |
+
transforms.Resize((224, 224)),
|
152 |
+
transforms.ToTensor(),
|
153 |
+
])
|
154 |
+
|
155 |
+
source = transform(source) # Update the source image size to 224x224, 1 of 2 sizes accepted by Yolo classification model
|
156 |
+
|
157 |
+
#result = detection_model.predict(source, conf=0.55, iou=0.6)
|
158 |
+
result = detection_model.predict(source) # Make prediction
|
159 |
+
# Get the top prediction
|
160 |
+
label = result[0].probs.top1
|
161 |
+
|
162 |
+
class_names = detection_model.names # Retrieves the class names mapping (dict-like)
|
163 |
+
classified_type = class_names[label] # Map numeric label to class name
|
164 |
+
print (">>> Class : ", classified_type)
|
165 |
+
|
166 |
+
confidence = result[0].probs.top1conf # Get the top class confidence
|
167 |
+
print(">>> Confidence : ", confidence)
|
168 |
+
|
169 |
+
annotated_image = pilimg.convert("RGB")
|
170 |
+
draw = ImageDraw.Draw(annotated_image)
|
171 |
+
font = ImageFont.truetype("arialbd.ttf", 30) # Use arial.ttf for bold font
|
172 |
+
|
173 |
+
if classified_type == classes[0]:
|
174 |
+
draw.text((300, 10), classified_type, fill="red", font=font)
|
175 |
+
gr.Warning("Defect detected, BAD!.")
|
176 |
+
else:
|
177 |
+
draw.text((300, 10), classified_type, fill="green", font=font)
|
178 |
+
gr.Info("No defect detected,GOOD!")
|
179 |
+
|
180 |
+
#draw.text((300, 10), classified_type, fill="red", font=font)
|
181 |
+
|
182 |
+
return annotated_image
|
183 |
+
|
184 |
+
detection_model = load_model_local()
|
185 |
+
|
186 |
+
title = "Detect the status of the cap, DEFECTIVE or GOOD"
|
187 |
+
interface = gr.Interface(
|
188 |
+
fn=predict,
|
189 |
+
inputs=gr.Image(type="pil", label="Input Image"),
|
190 |
+
outputs=gr.Image(type="pil", label="Classification result"),
|
191 |
+
title=title,
|
192 |
+
)
|
193 |
+
|
194 |
+
# Launch the interface
|
195 |
+
interface.launch(share=True)
|
196 |
+
|
197 |
+
|
198 |
+
# In[ ]:
|
199 |
+
|
200 |
+
|
201 |
+
|
202 |
+
|
203 |
+
|
204 |
+
# In[ ]:
|
205 |
+
|
206 |
+
|
207 |
+
|
208 |
+
|
209 |
+
|
210 |
+
# In[ ]:
|
211 |
+
|
212 |
+
|
213 |
+
|
214 |
+
|