def _prepare_img_to_run(self, img):
assert self.nn.batch_size == 1, \
"batch_size of runner is not 1, but trying to run against 1 image"
img = get_image_matrix(img, show=False)
# shape could be (x, y) or (3, x, y)
assert img.shape in [self.nn.input_shape[1:], self.nn.input_shape[2:]]
return img.flatten()
def run_batch(self, imgs):
nr_img = len(imgs)
assert all([img.shape[:2] == self.input_shape for img in imgs])
def preprocess(img):
img = mean_subtract(img)
return img * 2.0 / 255 - 1
# to ic01
images_to_run = np.asarray(
[get_image_matrix(preprocess(m), False) for m in imgs])
results = []
nowid = 0
for k in reversed(self.BATCHES):
nn = self.nns[k]
while nr_img >= k:
print k
inputs = images_to_run[nowid:nowid + k]
inputs = inputs.reshape((k, -1))
outputs = nn.func(inputs)
results.append(outputs)
nr_img -= k
nowid += k
results = np.concatenate(results, axis=0)
return results
def _prepare_img_to_run(self, img):
assert self.nn.batch_size == 1, \
"batch_size of runner is not 1, but trying to run against 1 image"
img = get_image_matrix(img, show=False)
# shape could be (x, y) or (3, x, y)
assert img.shape in [self.nn.input_shape[1:], self.nn.input_shape[2:]]
return img.flatten()
def run_batch(self, imgs):
nr_img = len(imgs)
assert all([img.shape[:2] == self.input_shape for img in imgs])
def preprocess(img):
img = mean_subtract(img)
return img * 2.0 / 255 - 1
# to ic01
images_to_run = np.asarray(
[get_image_matrix(preprocess(m), False) for m in imgs])
results = []
nowid = 0
for k in reversed(self.BATCHES):
nn = self.nns[k]
while nr_img >= k:
print k
inputs = images_to_run[nowid:nowid+k]
inputs = inputs.reshape((k, -1))
outputs = nn.func(inputs)
results.append(outputs)
nr_img -= k
nowid += k
results = np.concatenate(results, axis=0)
return results