def predict(self, image):
boxes = get_all_boxes(image)
images = time.measure(lambda: get_input(image, boxes),
'image preprocessing')
result = time.measure(lambda: self.model.predict(images),
'localization')
result = non_max_suppression(result)
return result
def predict(self, image):
boxes = get_all_boxes(image)
images = time.measure(lambda: get_input(image, boxes),
'image preprocessing')
cls, reg = time.measure(lambda: self.model.predict(images),
'localization')
result = np.concatenate((cls[..., 1:], reg), axis=-1)
result = non_max_suppression(result)
return result
def predict_multiple(self, images):
boxes = [get_all_boxes(image) for image in images]
inputs = time.measure(lambda: get_inputs(images, boxes),
'image preprocessing')
cls, reg = time.measure(lambda: self.model.predict(inputs),
'localization')
results = np.reshape(np.concatenate((cls[..., 1:], reg), axis=-1),
(len(boxes), len(boxes[0]), 5))
results = [non_max_suppression(result) for result in results]
return results
def detect(self, image):
objects = time.measure(lambda: self.detector.predict(image),
'detection')
extend_bounding_boxes(objects, 0.15)
images = time.measure(
lambda: prepare_for_classification(objects, image),
'image preprocessing')
labels = time.measure(lambda: self.classifier.predict(images),
'classification')
print(objects, labels)
return objects, labels
def predict_multiple(self, images):
boxes = get_all_boxes()
inputs, preprocessed_images = time.measure(lambda: get_all_inputs(images, boxes), 'preprocess')
results = None
for i in range(0, 4):
cls, reg = time.measure(lambda: self.models[i].predict(inputs[i]), f'detection {i}')
cls, reg = np.asarray(cls), np.asarray(reg)
result = np.reshape(
np.concatenate((cls[..., 1:], reg), axis=-1),
(len(images), len(boxes[i]), 5)
)
if results is None:
results = result
else:
results = np.concatenate((results, result), axis=1)
results = [non_max_suppression(result) for result in results]
box_scales = np.asarray([[len(image[0]) / 256, len(image) / 256, len(image[0]) / 256, len(image) / 256] for image in images])
return results, preprocessed_images, box_scales
def detect(request):
body = json.parse(request)
path = body["path"]
print(path)
image = load_image(path)
print(len(image), len(image[0]))
objects, labels = time.measure(
lambda: detector.detect_multiple(np.asarray([image])),
'the whole process')
return HttpResponse(json.convert(objects[0], labels[0]),
content_type="application/json")
def detect_multiple(self, images):
objects, preprocessed_images, box_scales = time.measure(
lambda: self.detector.predict_multiple(images), 'detection')
if len(objects[0]) == 0:
return [[]], [[]]
preprocessed = time.measure(
lambda: resize_for_classification(objects, preprocessed_images,
images), 'preprocessing')
labels = time.measure(lambda: self.classifier.predict(preprocessed),
'classification')
results = []
j = 0
for i in range(0, len(images)):
results.append([labels[k] for k in range(j, j + len(objects[i]))])
j += len(objects[i])
r_objects = []
for i in range(0, len(objects)):
r_objects.append([])
objs = objects[i]
r_objects[i] = [[obj[0], *obj[1:] * box_scales[i]] for obj in objs]
print(r_objects, results)
return r_objects, results
import sys
from pathlib import Path
import numpy as np
from server.server import run
mode = sys.argv[1]
assert mode == 'runserver' or mode == 'local'
if mode == 'runserver':
run()
else:
from utils import load_image, time
from trafficsigndetector.traffic_sign_detector import TrafficSignDetector
image = load_image('{root}/assets/images/{image}.png'.format(
root=Path(__file__).parent, image='testimage'))
detector = TrafficSignDetector()
time.measure(
lambda: detector.detect_multiple(
np.asarray([image for _ in range(0, 1)])), 'whole process')