Last active
July 26, 2019 08:04
-
-
Save a2chub/5267b91a3c1f3c97845fd2219f0023f2 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
"""A demo which runs object detection on OpenCV frames.""" | |
import argparse | |
import cv2 | |
import logging | |
import os | |
import edgetpu.detection.engine | |
FONT = cv2.FONT_HERSHEY_SIMPLEX | |
FONT_SCALE = 1.0 | |
FONT_THICKNESS = 2 | |
FONT_COLOR = (0, 0, 255) | |
BOX_THICKNESS = 2 | |
BOX_COLOR = (255, 0, 0) | |
def labels(): | |
with open('models/coco_labels.txt') as f: | |
for l in f.read().split('\n'): | |
parts = l.split(' ') | |
yield int(parts[0]), ' '.join(parts[1:]).strip() | |
labels = dict(labels()) | |
print(labels) | |
def main(): | |
parser = argparse.ArgumentParser() | |
parser.add_argument('--model', help='.tflite model path.', required=True) | |
args = parser.parse_args() | |
video_capture = cv2.VideoCapture(1) | |
video_capture.set(cv2.CAP_PROP_FRAME_WIDTH, 640) | |
video_capture.set(cv2.CAP_PROP_FRAME_HEIGHT, 480) | |
engine = edgetpu.detection.engine.DetectionEngine(os.path.join(args.model)) | |
input_tensor_size = tuple(engine.get_input_tensor_shape()[1:3]) | |
while video_capture.isOpened(): | |
# Capture frame-by-frame. | |
ret, frame = video_capture.read() | |
if not ret: | |
logging.error('video capture failed') | |
break | |
frame = cv2.copyMakeBorder(frame, 0, 160, 0, 0, cv2.BORDER_CONSTANT, value=[0, 0, 0]) | |
# Convert to tensor input format. | |
input_tensor = cv2.cvtColor(cv2.resize(frame, input_tensor_size), | |
cv2.COLOR_BGR2RGB).flatten() | |
# Run Inference. | |
result = engine.DetectWithInputTensor(input_tensor, | |
threshold=0.5, top_k=10) | |
# Process results. | |
if result: | |
for obj in result: | |
# Draw bounding box in frame coordinate space. | |
frame_h, frame_w, _ = frame.shape | |
bbox = (obj.bounding_box.flatten() | |
* [frame_w, frame_h, frame_w, frame_h]).astype(int) | |
cv2.rectangle(frame, tuple(bbox[:2]), tuple(bbox[2:]), | |
BOX_COLOR, BOX_THICKNESS) | |
label = labels[obj.label_id] if obj.label_id in labels else str(obj.label_id) | |
cv2.putText(frame, '%s:%.2f' % (label, obj.score), tuple(bbox[:2]), FONT, FONT_SCALE, FONT_COLOR, FONT_THICKNESS) | |
# Display inference time. | |
text = '(q to exit) inference: {:>7.3f}ms'.format( | |
engine.get_inference_time()) | |
size, _ = cv2.getTextSize(text, FONT, FONT_SCALE, FONT_THICKNESS) | |
cv2.putText(frame, text, (0, size[1]), | |
FONT, FONT_SCALE, FONT_COLOR, FONT_THICKNESS) | |
# Display the resulting frame | |
cv2.imshow('Video', frame) | |
# Wait for keypress. | |
if cv2.waitKey(1) & 0xFF == ord('q'): | |
break | |
# Release the capture | |
video_capture.release() | |
cv2.destroyAllWindows() | |
if __name__ == '__main__': | |
main() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment