def predict_images():
config = Config(FLAGS.model_dir)
detector = Detector(FLAGS.model_dir,
config,
use_gpu=FLAGS.use_gpu,
run_mode=config.mode)
if FLAGS.run_benchmark:
detector.predict(FLAGS.image_file,
detector.config.draw_threshold,
warmup=10,
repeats=10)
else:
# if os.path.exists(FLAGS.output_dir): shutil.rmtree(FLAGS.output_dir)
# os.makedirs(FLAGS.output_dir)
if not os.path.exists(FLAGS.output_dir): os.makedirs(FLAGS.output_dir)
postprocess = config.postprocess
if postprocess == 'numpy_nms':
pcfg = PostprocessNumpyNMSConfig()
# 获取颜色
num_classes = len(detector.config.labels)
colors = get_colors(num_classes)
path_dir = os.listdir(FLAGS.image_dir)
# warm up
if FLAGS.use_gpu:
for k, filename in enumerate(path_dir):
img_path = FLAGS.image_dir + filename
if postprocess == 'fastnms':
results = detector.predict(img_path,
detector.config.draw_threshold)
elif postprocess == 'numpy_nms':
results = detector.predict_with_postprocess(
img_path, detector.config.draw_threshold, pcfg)
if k == 10:
break
num_imgs = len(path_dir)
start = time.time()
for k, filename in enumerate(path_dir):
img_path = FLAGS.image_dir + filename
if postprocess == 'fastnms':
results = detector.predict(img_path,
detector.config.draw_threshold)
elif postprocess == 'numpy_nms':
results = detector.predict_with_postprocess(
img_path, detector.config.draw_threshold, pcfg)
image = cv2.imread(img_path)
if len(results['boxes']) > 0:
draw(image, results['boxes'], results['scores'],
results['classes'], detector.config.labels, colors)
out_path = os.path.join(FLAGS.output_dir, filename)
cv2.imwrite(out_path, image)
print("Detection bbox results save in {}".format(out_path))
cost = time.time() - start
print('total time: {0:.6f}s'.format(cost))
print('Speed: %.6fs per image, %.1f FPS.' % ((cost / num_imgs),
(num_imgs / cost)))
def play_video():
config = Config(FLAGS.model_dir)
detector = Detector(FLAGS.model_dir,
config,
use_gpu=FLAGS.use_gpu,
run_mode=config.mode)
# if os.path.exists(FLAGS.output_dir): shutil.rmtree(FLAGS.output_dir)
# os.makedirs(FLAGS.output_dir)
if not os.path.exists(FLAGS.output_dir): os.makedirs(FLAGS.output_dir)
# 获取颜色
num_classes = len(detector.config.labels)
colors = get_colors(num_classes)
# warm up
image_dir = 'images/test/'
path_dir = os.listdir(image_dir)
if FLAGS.use_gpu:
for k, filename in enumerate(path_dir):
img_path = image_dir + filename
results = detector.predict(img_path,
detector.config.draw_threshold)
if k == 10:
break
capture = cv2.VideoCapture(FLAGS.play_video)
fps = 30
width = int(capture.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
video_name = os.path.split(FLAGS.play_video)[-1]
if not os.path.exists(FLAGS.output_dir):
os.makedirs(FLAGS.output_dir)
out_path = os.path.join(FLAGS.output_dir, video_name)
writer = cv2.VideoWriter(out_path, fourcc, fps, (width, height))
index = 1
start = time.time()
while (1):
ret, frame = capture.read()
if not ret:
break
print('detect frame:%d' % (index))
index += 1
results = detector.predict(frame, detector.config.draw_threshold)
im = visualize_box_mask(frame, results, detector.config.labels)
cv2.imshow("detection", frame)
writer.write(im)
if cv2.waitKey(110) & 0xff == 27:
break
writer.release()
num_imgs = 100
cost = time.time() - start
print('total time: {0:.6f}s'.format(cost))
print('Speed: %.6fs per image, %.1f FPS.' % ((cost / num_imgs),
(num_imgs / cost)))
from tools.cocotools import get_classes from tools.visualize import get_colors, draw from paddle_serving_client import Client import cv2 import sys import numpy as np classes_path = 'data/coco_classes.txt' all_classes = get_classes(classes_path) num_classes = len(all_classes) colors = get_colors(num_classes) prototxt_path = sys.argv[1] client = Client() client.load_client_config(prototxt_path) client.connect(['127.0.0.1:9494']) img_path = sys.argv[3] print(img_path) # 这是图片的路径 input_shape = (608, 608) image = cv2.imread(img_path) h, w = image.shape[:2] img = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)