Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,4 +1,3 @@
|
|
| 1 |
-
import gc
|
| 2 |
import cv2
|
| 3 |
import tempfile
|
| 4 |
import numpy as np
|
|
@@ -18,61 +17,80 @@ POSE_CONNECTIONS = [
|
|
| 18 |
(30, 32)
|
| 19 |
]
|
| 20 |
|
|
|
|
| 21 |
@st.cache_resource
|
| 22 |
def load_model():
|
| 23 |
return tf.saved_model.load("Models/ssd_mobilenet/saved_model")
|
| 24 |
|
|
|
|
| 25 |
model = load_model()
|
| 26 |
mp_pose = mp.solutions.pose
|
| 27 |
|
| 28 |
labels = {1: 'person'}
|
| 29 |
|
|
|
|
| 30 |
def detect_persons(image):
|
| 31 |
tensor_img = tf.convert_to_tensor(image)
|
| 32 |
tensor_img = tensor_img[tf.newaxis, ...]
|
|
|
|
| 33 |
detections = model(tensor_img)
|
|
|
|
| 34 |
boxes = detections['detection_boxes'][0].numpy()
|
| 35 |
scores = detections['detection_scores'][0].numpy()
|
| 36 |
classes = detections['detection_classes'][0].numpy().astype(np.int32)
|
|
|
|
| 37 |
return boxes, scores, classes
|
| 38 |
|
|
|
|
| 39 |
def draw_landmarks(img, landmarks):
|
| 40 |
height, width, _ = img.shape
|
| 41 |
for lm in landmarks.landmark:
|
| 42 |
cx, cy = int(lm.x * width), int(lm.y * height)
|
| 43 |
cv2.circle(img, (cx, cy), 8, (0, 0, 255), -1)
|
|
|
|
| 44 |
for connection in POSE_CONNECTIONS:
|
| 45 |
start_idx, end_idx = connection
|
| 46 |
if landmarks.landmark[start_idx] and landmarks.landmark[end_idx]:
|
| 47 |
start_point = landmarks.landmark[start_idx]
|
| 48 |
end_point = landmarks.landmark[end_idx]
|
|
|
|
| 49 |
start_coordinates = (int(start_point.x * width), int(start_point.y * height))
|
| 50 |
end_coordinates = (int(end_point.x * width), int(end_point.y * height))
|
|
|
|
| 51 |
cv2.line(img, start_coordinates, end_coordinates, (0, 255, 0), 3)
|
|
|
|
| 52 |
return img
|
| 53 |
|
|
|
|
| 54 |
def draw_bounding_box(img, box, width, height):
|
| 55 |
y_min, x_min, y_max, x_max = box
|
| 56 |
left, right, top, bottom = x_min * width, x_max * width, y_min * height, y_max * height
|
| 57 |
cv2.rectangle(img, (int(left), int(top)), (int(right), int(bottom)), (255, 0, 0), 2)
|
| 58 |
|
|
|
|
| 59 |
def process_frame(frame, pose, draw_box):
|
| 60 |
image_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
| 61 |
boxes, scores, classes = detect_persons(image_rgb)
|
|
|
|
| 62 |
height, width, _ = frame.shape
|
| 63 |
for i in range(len(scores)):
|
| 64 |
if scores[i] > 0.5 and classes[i] == 1:
|
| 65 |
y_min, x_min, y_max, x_max = boxes[i]
|
| 66 |
left, right, top, bottom = x_min * width, x_max * width, y_min * height, y_max * height
|
| 67 |
person_roi = frame[int(top):int(bottom), int(left):int(right)]
|
|
|
|
| 68 |
results = pose.process(cv2.cvtColor(person_roi, cv2.COLOR_BGR2RGB))
|
|
|
|
| 69 |
if results.pose_landmarks:
|
| 70 |
person_roi = draw_landmarks(person_roi, results.pose_landmarks)
|
|
|
|
| 71 |
frame[int(top):int(bottom), int(left):int(right)] = person_roi
|
| 72 |
if draw_box:
|
| 73 |
draw_bounding_box(frame, boxes[i], width, height)
|
|
|
|
| 74 |
return frame
|
| 75 |
|
|
|
|
| 76 |
def main():
|
| 77 |
st.markdown(
|
| 78 |
"""
|
|
@@ -153,9 +171,7 @@ def main():
|
|
| 153 |
frame = process_frame(frame, pose, draw_box)
|
| 154 |
|
| 155 |
st_frame.image(frame, channels='BGR', use_column_width=True)
|
| 156 |
-
|
| 157 |
-
# Ensure proper synchronization and frame display
|
| 158 |
-
st_frame.empty()
|
| 159 |
|
| 160 |
st.text("Completed")
|
| 161 |
cam.release()
|
|
@@ -166,8 +182,6 @@ def main():
|
|
| 166 |
|
| 167 |
st.image(processed_image, channels='BGR', use_column_width=True)
|
| 168 |
|
| 169 |
-
gc.collect()
|
| 170 |
-
|
| 171 |
elif operation_type == "Demo":
|
| 172 |
st.empty()
|
| 173 |
st.markdown("<p class='intro'>Demo video will be shown below:</p>", unsafe_allow_html=True)
|
|
@@ -185,11 +199,11 @@ def main():
|
|
| 185 |
frame = process_frame(frame, pose, draw_box=False)
|
| 186 |
|
| 187 |
st_frame.image(frame, channels='BGR', use_column_width=True)
|
| 188 |
-
|
| 189 |
|
| 190 |
st.text("Completed")
|
| 191 |
cam.release()
|
| 192 |
-
|
| 193 |
|
| 194 |
if __name__ == "__main__":
|
| 195 |
main()
|
|
|
|
|
|
|
| 1 |
import cv2
|
| 2 |
import tempfile
|
| 3 |
import numpy as np
|
|
|
|
| 17 |
(30, 32)
|
| 18 |
]
|
| 19 |
|
| 20 |
+
|
| 21 |
@st.cache_resource
|
| 22 |
def load_model():
|
| 23 |
return tf.saved_model.load("Models/ssd_mobilenet/saved_model")
|
| 24 |
|
| 25 |
+
|
| 26 |
model = load_model()
|
| 27 |
mp_pose = mp.solutions.pose
|
| 28 |
|
| 29 |
labels = {1: 'person'}
|
| 30 |
|
| 31 |
+
|
| 32 |
def detect_persons(image):
|
| 33 |
tensor_img = tf.convert_to_tensor(image)
|
| 34 |
tensor_img = tensor_img[tf.newaxis, ...]
|
| 35 |
+
|
| 36 |
detections = model(tensor_img)
|
| 37 |
+
|
| 38 |
boxes = detections['detection_boxes'][0].numpy()
|
| 39 |
scores = detections['detection_scores'][0].numpy()
|
| 40 |
classes = detections['detection_classes'][0].numpy().astype(np.int32)
|
| 41 |
+
|
| 42 |
return boxes, scores, classes
|
| 43 |
|
| 44 |
+
|
| 45 |
def draw_landmarks(img, landmarks):
|
| 46 |
height, width, _ = img.shape
|
| 47 |
for lm in landmarks.landmark:
|
| 48 |
cx, cy = int(lm.x * width), int(lm.y * height)
|
| 49 |
cv2.circle(img, (cx, cy), 8, (0, 0, 255), -1)
|
| 50 |
+
|
| 51 |
for connection in POSE_CONNECTIONS:
|
| 52 |
start_idx, end_idx = connection
|
| 53 |
if landmarks.landmark[start_idx] and landmarks.landmark[end_idx]:
|
| 54 |
start_point = landmarks.landmark[start_idx]
|
| 55 |
end_point = landmarks.landmark[end_idx]
|
| 56 |
+
|
| 57 |
start_coordinates = (int(start_point.x * width), int(start_point.y * height))
|
| 58 |
end_coordinates = (int(end_point.x * width), int(end_point.y * height))
|
| 59 |
+
|
| 60 |
cv2.line(img, start_coordinates, end_coordinates, (0, 255, 0), 3)
|
| 61 |
+
|
| 62 |
return img
|
| 63 |
|
| 64 |
+
|
| 65 |
def draw_bounding_box(img, box, width, height):
|
| 66 |
y_min, x_min, y_max, x_max = box
|
| 67 |
left, right, top, bottom = x_min * width, x_max * width, y_min * height, y_max * height
|
| 68 |
cv2.rectangle(img, (int(left), int(top)), (int(right), int(bottom)), (255, 0, 0), 2)
|
| 69 |
|
| 70 |
+
|
| 71 |
def process_frame(frame, pose, draw_box):
|
| 72 |
image_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
| 73 |
boxes, scores, classes = detect_persons(image_rgb)
|
| 74 |
+
|
| 75 |
height, width, _ = frame.shape
|
| 76 |
for i in range(len(scores)):
|
| 77 |
if scores[i] > 0.5 and classes[i] == 1:
|
| 78 |
y_min, x_min, y_max, x_max = boxes[i]
|
| 79 |
left, right, top, bottom = x_min * width, x_max * width, y_min * height, y_max * height
|
| 80 |
person_roi = frame[int(top):int(bottom), int(left):int(right)]
|
| 81 |
+
|
| 82 |
results = pose.process(cv2.cvtColor(person_roi, cv2.COLOR_BGR2RGB))
|
| 83 |
+
|
| 84 |
if results.pose_landmarks:
|
| 85 |
person_roi = draw_landmarks(person_roi, results.pose_landmarks)
|
| 86 |
+
|
| 87 |
frame[int(top):int(bottom), int(left):int(right)] = person_roi
|
| 88 |
if draw_box:
|
| 89 |
draw_bounding_box(frame, boxes[i], width, height)
|
| 90 |
+
|
| 91 |
return frame
|
| 92 |
|
| 93 |
+
|
| 94 |
def main():
|
| 95 |
st.markdown(
|
| 96 |
"""
|
|
|
|
| 171 |
frame = process_frame(frame, pose, draw_box)
|
| 172 |
|
| 173 |
st_frame.image(frame, channels='BGR', use_column_width=True)
|
| 174 |
+
st.empty()
|
|
|
|
|
|
|
| 175 |
|
| 176 |
st.text("Completed")
|
| 177 |
cam.release()
|
|
|
|
| 182 |
|
| 183 |
st.image(processed_image, channels='BGR', use_column_width=True)
|
| 184 |
|
|
|
|
|
|
|
| 185 |
elif operation_type == "Demo":
|
| 186 |
st.empty()
|
| 187 |
st.markdown("<p class='intro'>Demo video will be shown below:</p>", unsafe_allow_html=True)
|
|
|
|
| 199 |
frame = process_frame(frame, pose, draw_box=False)
|
| 200 |
|
| 201 |
st_frame.image(frame, channels='BGR', use_column_width=True)
|
| 202 |
+
st.empty()
|
| 203 |
|
| 204 |
st.text("Completed")
|
| 205 |
cam.release()
|
| 206 |
+
|
| 207 |
|
| 208 |
if __name__ == "__main__":
|
| 209 |
main()
|