Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -7,8 +7,20 @@ from PIL import Image
|
|
7 |
|
8 |
# Load the TensorFlow Lite model
|
9 |
MODEL_DIR = 'model'
|
10 |
-
|
11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
|
13 |
pkg = importlib.util.find_spec('tflite_runtime')
|
14 |
if pkg:
|
@@ -18,80 +30,87 @@ else:
|
|
18 |
from tensorflow.lite.python.interpreter import Interpreter
|
19 |
from tensorflow.lite.python.interpreter import load_delegate
|
20 |
|
21 |
-
PATH_TO_CKPT = os.path.join(MODEL_DIR, GRAPH_NAME)
|
22 |
-
PATH_TO_LABELS = os.path.join(MODEL_DIR, LABELMAP_NAME)
|
23 |
-
|
24 |
# Load the label map
|
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 |
def resize_image(image, size=640):
|
86 |
return cv2.resize(image, (size, size))
|
87 |
|
88 |
-
def detect_image(input_image):
|
89 |
image = np.array(input_image)
|
90 |
resized_image = resize_image(image, size=640) # Resize input image
|
91 |
-
result_image = perform_detection(resized_image
|
92 |
return Image.fromarray(result_image)
|
93 |
|
94 |
-
def detect_video(input_video):
|
95 |
cap = cv2.VideoCapture(input_video)
|
96 |
frames = []
|
97 |
|
@@ -101,7 +120,7 @@ def detect_video(input_video):
|
|
101 |
break
|
102 |
|
103 |
resized_frame = resize_image(frame, size=640) # Resize each frame
|
104 |
-
result_frame = perform_detection(resized_frame
|
105 |
frames.append(result_frame)
|
106 |
|
107 |
cap.release()
|
@@ -124,17 +143,19 @@ def detect_video(input_video):
|
|
124 |
app = gr.Blocks()
|
125 |
|
126 |
with app:
|
|
|
|
|
|
|
127 |
with gr.Tab("Image Detection"):
|
128 |
gr.Markdown("Upload an image for object detection")
|
129 |
image_input = gr.Image(type="pil", label="Upload an image")
|
130 |
image_output = gr.Image(type="pil", label="Detection Result")
|
131 |
-
gr.Button("Submit").click(fn=detect_image, inputs=image_input, outputs=image_output)
|
132 |
|
133 |
with gr.Tab("Video Detection"):
|
134 |
gr.Markdown("Upload a video for object detection")
|
135 |
video_input = gr.Video(label="Upload a video")
|
136 |
video_output = gr.Video(label="Detection Result")
|
137 |
-
gr.Button("Submit").click(fn=detect_video, inputs=video_input, outputs=video_output)
|
138 |
|
139 |
app.launch()
|
140 |
-
|
|
|
7 |
|
8 |
# Load the TensorFlow Lite model
|
9 |
MODEL_DIR = 'model'
|
10 |
+
MODEL_DIRS = {
|
11 |
+
'Multi-class model': 'model',
|
12 |
+
'Empty class': 'model_2',
|
13 |
+
'Misalignment class': 'model_3'
|
14 |
+
}
|
15 |
+
|
16 |
+
# Function to load model based on selection
|
17 |
+
def load_model(model_name):
|
18 |
+
selected_model_dir = MODEL_DIRS.get(model_name, MODEL_DIR)
|
19 |
+
graph_name = 'detect.tflite' if model_name == 'Multi-class model' else f'detect_{model_name.lower().replace(" ", "_")}.tflite'
|
20 |
+
labelmap_name = 'labelmap.txt' if model_name == 'Multi-class model' else f'labelmap_{model_name.lower().replace(" ", "_")}.txt'
|
21 |
+
path_to_ckpt = os.path.join(selected_model_dir, graph_name)
|
22 |
+
path_to_labels = os.path.join(selected_model_dir, labelmap_name)
|
23 |
+
return path_to_ckpt, path_to_labels
|
24 |
|
25 |
pkg = importlib.util.find_spec('tflite_runtime')
|
26 |
if pkg:
|
|
|
30 |
from tensorflow.lite.python.interpreter import Interpreter
|
31 |
from tensorflow.lite.python.interpreter import load_delegate
|
32 |
|
|
|
|
|
|
|
33 |
# Load the label map
|
34 |
+
def load_labels(path_to_labels):
|
35 |
+
with open(path_to_labels, 'r') as f:
|
36 |
+
labels = [line.strip() for line in f.readlines()]
|
37 |
+
|
38 |
+
if labels[0] == '???':
|
39 |
+
del(labels[0])
|
40 |
+
|
41 |
+
return labels
|
42 |
+
|
43 |
+
def load_interpreter(model_path):
|
44 |
+
interpreter = Interpreter(model_path=model_path)
|
45 |
+
interpreter.allocate_tensors()
|
46 |
+
return interpreter
|
47 |
+
|
48 |
+
class ModelDetector:
|
49 |
+
def __init__(self, model_name):
|
50 |
+
self.model_path, self.label_path = load_model(model_name)
|
51 |
+
self.labels = load_labels(self.label_path)
|
52 |
+
self.interpreter = load_interpreter(self.model_path)
|
53 |
+
|
54 |
+
input_details = self.interpreter.get_input_details()
|
55 |
+
output_details = self.interpreter.get_output_details()
|
56 |
+
self.height = input_details[0]['shape'][1]
|
57 |
+
self.width = input_details[0]['shape'][2]
|
58 |
+
self.floating_model = (input_details[0]['dtype'] == np.float32)
|
59 |
+
|
60 |
+
self.input_mean = 127.5
|
61 |
+
self.input_std = 127.5
|
62 |
+
|
63 |
+
outname = output_details[0]['name']
|
64 |
+
if ('StatefulPartitionedCall' in outname):
|
65 |
+
self.boxes_idx, self.classes_idx, self.scores_idx = 1, 3, 0
|
66 |
+
else:
|
67 |
+
self.boxes_idx, self.classes_idx, self.scores_idx = 0, 1, 2
|
68 |
+
|
69 |
+
def perform_detection(self, image):
|
70 |
+
imH, imW, _ = image.shape
|
71 |
+
image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
|
72 |
+
image_resized = cv2.resize(image_rgb, (self.width, self.height))
|
73 |
+
input_data = np.expand_dims(image_resized, axis=0)
|
74 |
+
|
75 |
+
if self.floating_model:
|
76 |
+
input_data = (np.float32(input_data) - self.input_mean) / self.input_std
|
77 |
+
|
78 |
+
self.interpreter.set_tensor(self.interpreter.get_input_details()[0]['index'], input_data)
|
79 |
+
self.interpreter.invoke()
|
80 |
+
|
81 |
+
boxes = self.interpreter.get_tensor(self.interpreter.get_output_details()[self.boxes_idx]['index'])[0]
|
82 |
+
classes = self.interpreter.get_tensor(self.interpreter.get_output_details()[self.classes_idx]['index'])[0]
|
83 |
+
scores = self.interpreter.get_tensor(self.interpreter.get_output_details()[self.scores_idx]['index'])[0]
|
84 |
+
|
85 |
+
detections = []
|
86 |
+
for i in range(len(scores)):
|
87 |
+
if ((scores[i] > 0.5) and (scores[i] <= 1.0)):
|
88 |
+
ymin = int(max(1, (boxes[i][0] * imH)))
|
89 |
+
xmin = int(max(1, (boxes[i][1] * imW)))
|
90 |
+
ymax = int(min(imH, (boxes[i][2] * imH)))
|
91 |
+
xmax = int(min(imW, (boxes[i][3] * imW)))
|
92 |
+
|
93 |
+
cv2.rectangle(image, (xmin, ymin), (xmax, ymax), (10, 255, 0), 2)
|
94 |
+
object_name = self.labels[int(classes[i])]
|
95 |
+
label = '%s: %d%%' % (object_name, int(scores[i] * 100))
|
96 |
+
labelSize, baseLine = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.7, 2)
|
97 |
+
label_ymin = max(ymin, labelSize[1] + 10)
|
98 |
+
cv2.rectangle(image, (xmin, label_ymin - labelSize[1] - 10), (xmin + labelSize[0], label_ymin + baseLine - 10), (255, 255, 255), cv2.FILLED)
|
99 |
+
cv2.putText(image, label, (xmin, label_ymin - 7), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 0), 2)
|
100 |
+
|
101 |
+
detections.append([object_name, scores[i], xmin, ymin, xmax, ymax])
|
102 |
+
return image
|
103 |
|
104 |
def resize_image(image, size=640):
|
105 |
return cv2.resize(image, (size, size))
|
106 |
|
107 |
+
def detect_image(input_image, model_detector):
|
108 |
image = np.array(input_image)
|
109 |
resized_image = resize_image(image, size=640) # Resize input image
|
110 |
+
result_image = model_detector.perform_detection(resized_image)
|
111 |
return Image.fromarray(result_image)
|
112 |
|
113 |
+
def detect_video(input_video, model_detector):
|
114 |
cap = cv2.VideoCapture(input_video)
|
115 |
frames = []
|
116 |
|
|
|
120 |
break
|
121 |
|
122 |
resized_frame = resize_image(frame, size=640) # Resize each frame
|
123 |
+
result_frame = model_detector.perform_detection(resized_frame)
|
124 |
frames.append(result_frame)
|
125 |
|
126 |
cap.release()
|
|
|
143 |
app = gr.Blocks()
|
144 |
|
145 |
with app:
|
146 |
+
gr.Label("Select Model:")
|
147 |
+
model_selector = gr.Dropdown(choices=list(MODEL_DIRS.keys()), label="Multi-class model")
|
148 |
+
|
149 |
with gr.Tab("Image Detection"):
|
150 |
gr.Markdown("Upload an image for object detection")
|
151 |
image_input = gr.Image(type="pil", label="Upload an image")
|
152 |
image_output = gr.Image(type="pil", label="Detection Result")
|
153 |
+
gr.Button("Submit").click(fn=detect_image, inputs=[image_input, model_selector], outputs=image_output)
|
154 |
|
155 |
with gr.Tab("Video Detection"):
|
156 |
gr.Markdown("Upload a video for object detection")
|
157 |
video_input = gr.Video(label="Upload a video")
|
158 |
video_output = gr.Video(label="Detection Result")
|
159 |
+
gr.Button("Submit").click(fn=detect_video, inputs=[video_input, model_selector], outputs=video_output)
|
160 |
|
161 |
app.launch()
|
|