def __init__(self, network, settings):
self._settings = settings
self._input_shape = get_image_shape(network)
if self._settings.rescale_type is None:
net_meta = network["metadata"]
self._settings.rescale_type = net_meta.get("rescale_type", WARP)
def reader_for_network(network_name, additional_settings):
extras = dict(EXTRA_PARAMS)
if additional_settings:
extras.update(additional_settings)
image_shape = get_image_shape(get_pretrained_network(network_name))
return make_image_reader(Settings(extras), image_shape, False)
def reader_for_network(network_name, additional_settings):
extras = dict(EXTRA_PARAMS)
if additional_settings:
extras.update(additional_settings)
settings = Settings(extras)
image_shape = get_image_shape(get_pretrained_network(network_name))
return make_image_reader("file", image_shape, TEST_IMAGE_DATA, settings)
def __init__(self, network, nclasses):
self._nclasses = nclasses
self._input_size = get_image_shape(network)[1]
self._layers = network["layers"]
self._branches = []
for branch in yolo_outputs(network):
d_info = [branch[k] for k in ["strides", "anchors", "xyscale"]]
self._branches.append((branch["input"], d_info))
def __init__(self, network, nclasses, settings):
self._nclasses = nclasses
self._input_shape = get_image_shape(network)[1:3]
ob = yolo_outputs(network)
self._strides = tuple(
[self._input_shape[0] // b["strides"] for b in ob])
self._nanchors = tuple([len(b["anchors"]) for b in ob])
self._unfiltered = settings.output_unfiltered_boxes
self._threshold = settings.bounding_box_threshold or SCORE_THRESHOLD
self._iou_threshold = settings.iou_threshold or IOU_THRESHOLD
self._max_objects = settings.max_objects or MAX_OBJECTS
def __init__(self, network, nclasses):
super(YoloBranches, self).__init__()
self._nclasses = nclasses
self._input_size = get_image_shape(network)[1]
self._branches = []
assert network['layers'][-1]['type'] == 'yolo_output_branches'
out_branches = network['layers'][-1]
for i, branch in enumerate(out_branches['output_branches']):
idx = branch['input']
d_info = [branch[k] for k in ['strides', 'anchors', 'xyscale']]
layers = make_sequence(branch['convolution_path'], LAYER_FUNCTIONS)
self._branches.append((idx, d_info, layers))
def __init__(self, network, nclasses, settings):
super(BoxLocator, self).__init__()
assert network['layers'][-1]['type'] == 'yolo_output_branches'
self._nclasses = nclasses
self._input_shape = get_image_shape(network)[1:3]
ob = network['layers'][-1]['output_branches']
self._strides = tuple(
[self._input_shape[0] // b['strides'] for b in ob])
self._nanchors = tuple([len(b['anchors']) for b in ob])
self._unfiltered = settings.output_unfiltered_boxes
self._threshold = settings.bounding_box_threshold or SCORE_THRESHOLD
self._iou_threshold = settings.iou_threshold or IOU_THRESHOLD
self._max_objects = settings.max_objects or MAX_OBJECTS
def __init__(self, network, settings):
super(ImageReader, self).__init__()
self._input_shape = get_image_shape(network)
self._settings = settings