app file
Browse files
app.py
ADDED
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from flask import Flask, render_template, request, jsonify
|
2 |
+
import os
|
3 |
+
import cv2
|
4 |
+
import numpy as np
|
5 |
+
import tensorflow as tf
|
6 |
+
from object_detection.utils import label_map_util
|
7 |
+
from object_detection.utils import visualization_utils as viz_utils
|
8 |
+
from object_detection.builders import model_builder
|
9 |
+
from object_detection.utils import config_util
|
10 |
+
|
11 |
+
app = Flask(__name__)
|
12 |
+
|
13 |
+
# Load model and label map
|
14 |
+
CUSTOM_MODEL_NAME = 'my_ssd_mobnet'
|
15 |
+
paths = {
|
16 |
+
'CHECKPOINT_PATH': os.path.join('Tensorflow', 'workspace', 'models', CUSTOM_MODEL_NAME),
|
17 |
+
'LABELMAP': os.path.join('Tensorflow', 'workspace', 'annotations', 'label_map.pbtxt')
|
18 |
+
}
|
19 |
+
|
20 |
+
configs = config_util.get_configs_from_pipeline_file(os.path.join(paths['CHECKPOINT_PATH'], 'pipeline.config'))
|
21 |
+
detection_model = model_builder.build(model_config=configs['model'], is_training=False)
|
22 |
+
ckpt = tf.compat.v2.train.Checkpoint(model=detection_model)
|
23 |
+
ckpt.restore(os.path.join(paths['CHECKPOINT_PATH'], 'ckpt-7')).expect_partial()
|
24 |
+
category_index = label_map_util.create_category_index_from_labelmap(paths['LABELMAP'])
|
25 |
+
|
26 |
+
# Define detection function
|
27 |
+
@tf.function
|
28 |
+
def detect_fn(image):
|
29 |
+
image, shapes = detection_model.preprocess(image)
|
30 |
+
prediction_dict = detection_model.predict(image, shapes)
|
31 |
+
detections = detection_model.postprocess(prediction_dict, shapes)
|
32 |
+
return detections
|
33 |
+
|
34 |
+
# Define route for object detection
|
35 |
+
@app.route('/detect', methods=['POST'])
|
36 |
+
def detect():
|
37 |
+
# Get image file from request
|
38 |
+
file = request.files['image']
|
39 |
+
|
40 |
+
# Read image and convert to numpy array
|
41 |
+
img = cv2.imdecode(np.frombuffer(file.read(), np.uint8), cv2.IMREAD_COLOR)
|
42 |
+
image_np = np.array(img)
|
43 |
+
|
44 |
+
# Perform object detection
|
45 |
+
input_tensor = tf.convert_to_tensor(np.expand_dims(image_np, 0), dtype=tf.float32)
|
46 |
+
detections = detect_fn(input_tensor)
|
47 |
+
|
48 |
+
num_detections = int(detections.pop('num_detections'))
|
49 |
+
detections = {key: value[0, :num_detections].numpy() for key, value in detections.items()}
|
50 |
+
detections['num_detections'] = num_detections
|
51 |
+
detections['detection_classes'] = detections['detection_classes'].astype(np.int64)
|
52 |
+
|
53 |
+
label_id_offset = 1
|
54 |
+
image_np_with_detections = image_np.copy()
|
55 |
+
|
56 |
+
viz_utils.visualize_boxes_and_labels_on_image_array(
|
57 |
+
image_np_with_detections,
|
58 |
+
detections['detection_boxes'],
|
59 |
+
detections['detection_classes'] + label_id_offset,
|
60 |
+
detections['detection_scores'],
|
61 |
+
category_index,
|
62 |
+
use_normalized_coordinates=True,
|
63 |
+
max_boxes_to_draw=10,
|
64 |
+
min_score_thresh=.4,
|
65 |
+
agnostic_mode=False
|
66 |
+
)
|
67 |
+
|
68 |
+
# Convert image back to byte stream
|
69 |
+
ret, buffer = cv2.imencode('.jpg', image_np_with_detections)
|
70 |
+
img_str = buffer.tobytes()
|
71 |
+
|
72 |
+
return img_str
|
73 |
+
|
74 |
+
# Define index route
|
75 |
+
@app.route('/', methods=['GET'])
|
76 |
+
def index():
|
77 |
+
return render_template('index.html')
|
78 |
+
|
79 |
+
if __name__ == "__main__":
|
80 |
+
app.run(debug=True)
|