Update app.py
Browse files
app.py
CHANGED
@@ -16,31 +16,22 @@ PATH_TO_LABELS = 'data/label_map.pbtxt'
|
|
16 |
category_index = label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS, use_display_name=True)
|
17 |
|
18 |
def pil_image_as_numpy_array(pilimg):
|
19 |
-
|
20 |
img_array = tf.keras.utils.img_to_array(pilimg)
|
21 |
-
# img_array = np.expand_dims(img_array, axis=0)
|
22 |
return img_array
|
23 |
-
|
24 |
|
25 |
def load_model():
|
26 |
model_dir = 'saved_model'
|
27 |
detection_model = tf.saved_model.load(str(model_dir))
|
28 |
-
return detection_model
|
29 |
-
|
30 |
|
31 |
def predict(image_np):
|
32 |
-
|
33 |
image_np = pil_image_as_numpy_array(image_np)
|
34 |
image_np = np.expand_dims(image_np, axis=0)
|
35 |
-
|
36 |
results = detection_model(image_np)
|
37 |
-
|
38 |
-
# different object detection models have additional results
|
39 |
-
result = {key:value.numpy() for key,value in results.items()}
|
40 |
-
|
41 |
label_id_offset = 0
|
42 |
image_np_with_detections = image_np.copy()
|
43 |
-
|
44 |
viz_utils.visualize_boxes_and_labels_on_image_array(
|
45 |
image_np_with_detections[0],
|
46 |
result['detection_boxes'][0],
|
@@ -51,57 +42,47 @@ def predict(image_np):
|
|
51 |
max_boxes_to_draw=200,
|
52 |
min_score_thresh=.60,
|
53 |
agnostic_mode=False,
|
54 |
-
line_thickness=2
|
55 |
-
|
56 |
result_pil_img = tf.keras.utils.array_to_img(image_np_with_detections[0])
|
57 |
-
|
58 |
return result_pil_img
|
59 |
|
60 |
-
|
61 |
detection_model = load_model()
|
62 |
|
63 |
# Specify paths to example images
|
64 |
-
sample_images = [
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
|
|
|
|
|
|
73 |
|
74 |
def predict_on_video(video_in_filepath, video_out_filepath, detection_model, category_index):
|
|
|
75 |
video_reader = cv2.VideoCapture(video_in_filepath)
|
76 |
-
|
77 |
frame_h = int(video_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
78 |
frame_w = int(video_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
|
79 |
fps = video_reader.get(cv2.CAP_PROP_FPS)
|
80 |
-
|
81 |
video_writer = cv2.VideoWriter(
|
82 |
video_out_filepath,
|
83 |
cv2.VideoWriter_fourcc(*'mp4v'),
|
84 |
fps,
|
85 |
(frame_w, frame_h)
|
86 |
)
|
87 |
-
|
88 |
label_id_offset = 0
|
89 |
-
|
90 |
while True:
|
91 |
ret, frame = video_reader.read()
|
92 |
-
|
93 |
if not ret:
|
94 |
break # Break the loop if the video is finished
|
95 |
-
|
96 |
-
processed_frame = predict(frame, detection_model, category_index, label_id_offset)
|
97 |
-
|
98 |
-
# Convert processed frame to numpy array
|
99 |
processed_frame_np = np.array(processed_frame)
|
100 |
-
|
101 |
-
# Write the frame to the output video
|
102 |
video_writer.write(processed_frame_np)
|
103 |
-
|
104 |
-
# Release video reader and writer
|
105 |
video_reader.release()
|
106 |
video_writer.release()
|
107 |
cv2.destroyAllWindows()
|
@@ -110,7 +91,6 @@ def predict_on_video(video_in_filepath, video_out_filepath, detection_model, cat
|
|
110 |
# Function to process a video
|
111 |
def process_video(video_path):
|
112 |
output_path = "output_video.mp4" # Output path for the processed video
|
113 |
-
# Assuming you have detection_model and category_index defined
|
114 |
predict_on_video(video_path, output_path, detection_model, category_index)
|
115 |
return output_path
|
116 |
|
|
|
16 |
category_index = label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS, use_display_name=True)
|
17 |
|
18 |
def pil_image_as_numpy_array(pilimg):
|
|
|
19 |
img_array = tf.keras.utils.img_to_array(pilimg)
|
|
|
20 |
return img_array
|
|
|
21 |
|
22 |
def load_model():
|
23 |
model_dir = 'saved_model'
|
24 |
detection_model = tf.saved_model.load(str(model_dir))
|
25 |
+
return detection_model
|
|
|
26 |
|
27 |
def predict(image_np):
|
28 |
+
global detection_model # Declare as a global variable
|
29 |
image_np = pil_image_as_numpy_array(image_np)
|
30 |
image_np = np.expand_dims(image_np, axis=0)
|
|
|
31 |
results = detection_model(image_np)
|
32 |
+
result = {key: value.numpy() for key, value in results.items()}
|
|
|
|
|
|
|
33 |
label_id_offset = 0
|
34 |
image_np_with_detections = image_np.copy()
|
|
|
35 |
viz_utils.visualize_boxes_and_labels_on_image_array(
|
36 |
image_np_with_detections[0],
|
37 |
result['detection_boxes'][0],
|
|
|
42 |
max_boxes_to_draw=200,
|
43 |
min_score_thresh=.60,
|
44 |
agnostic_mode=False,
|
45 |
+
line_thickness=2
|
46 |
+
)
|
47 |
result_pil_img = tf.keras.utils.array_to_img(image_np_with_detections[0])
|
|
|
48 |
return result_pil_img
|
49 |
|
|
|
50 |
detection_model = load_model()
|
51 |
|
52 |
# Specify paths to example images
|
53 |
+
sample_images = [
|
54 |
+
["test_1.jpg"], ["test_9.jpg"], ["test_6.jpg"],
|
55 |
+
["test_7.jpg"], ["test_10.jpg"], ["test_11.jpg"], ["test_8.jpg"]
|
56 |
+
]
|
57 |
+
|
58 |
+
tab1 = gr.Interface(
|
59 |
+
fn=predict,
|
60 |
+
inputs=gr.Image(label='Upload an expressway image', type="pil"),
|
61 |
+
outputs=gr.Image(type="pil"),
|
62 |
+
title='Blue and Yellow Taxi detection in live expressway traffic conditions (data.gov.sg)',
|
63 |
+
examples=sample_images
|
64 |
+
)
|
65 |
|
66 |
def predict_on_video(video_in_filepath, video_out_filepath, detection_model, category_index):
|
67 |
+
global detection_model # Declare as a global variable
|
68 |
video_reader = cv2.VideoCapture(video_in_filepath)
|
|
|
69 |
frame_h = int(video_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
70 |
frame_w = int(video_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
|
71 |
fps = video_reader.get(cv2.CAP_PROP_FPS)
|
|
|
72 |
video_writer = cv2.VideoWriter(
|
73 |
video_out_filepath,
|
74 |
cv2.VideoWriter_fourcc(*'mp4v'),
|
75 |
fps,
|
76 |
(frame_w, frame_h)
|
77 |
)
|
|
|
78 |
label_id_offset = 0
|
|
|
79 |
while True:
|
80 |
ret, frame = video_reader.read()
|
|
|
81 |
if not ret:
|
82 |
break # Break the loop if the video is finished
|
83 |
+
processed_frame = predict(frame)
|
|
|
|
|
|
|
84 |
processed_frame_np = np.array(processed_frame)
|
|
|
|
|
85 |
video_writer.write(processed_frame_np)
|
|
|
|
|
86 |
video_reader.release()
|
87 |
video_writer.release()
|
88 |
cv2.destroyAllWindows()
|
|
|
91 |
# Function to process a video
|
92 |
def process_video(video_path):
|
93 |
output_path = "output_video.mp4" # Output path for the processed video
|
|
|
94 |
predict_on_video(video_path, output_path, detection_model, category_index)
|
95 |
return output_path
|
96 |
|