def test_check_valid_attributes():
attr_dict = {
"mode": "bilinear",
"coefficients": [1, 2, 3, 4, 5],
}
def _is_supported_mode(x):
return x in ["linear", "area", "cubic", "bilinear"]
requirements = [
("width", False, np.unsignedinteger, None),
("mode", True, np.str_, _is_supported_mode),
("coefficients", True, np.integer, lambda x: x > 0),
("alpha", False, np.float64, None),
]
assert check_valid_attributes("TestOp", attr_dict, requirements)
requirements[3] = ("alpha", True, np.float64, None)
try:
check_valid_attributes("TestOp", attr_dict, requirements)
except UserInputError:
pass
else:
raise AssertionError(
"Validation of missing required attribute has unexpectedly passed."
)
def detection_output(
box_logits: NodeInput,
class_preds: NodeInput,
proposals: NodeInput,
attrs: dict,
aux_class_preds: Optional[NodeInput] = None,
aux_box_preds: Optional[NodeInput] = None,
name: Optional[str] = None,
) -> Node:
"""Generate the detection output using information on location and confidence predictions.
@param box_logits: The 2D input tensor with box logits.
@param class_preds: The 2D input tensor with class predictions.
@param proposals: The 3D input tensor with proposals.
@param attrs: The dictionary containing key, value pairs for attributes.
@param aux_class_preds: The 2D input tensor with additional class predictions information.
@param aux_box_preds: The 2D input tensor with additional box predictions information.
@param name: Optional name for the output node.
@return Node representing DetectionOutput operation.
Available attributes are:
* background_label_id The background label id.
Range of values: integer value
Default value: 0
Required: no
* top_k Maximum number of results to be kept per batch after NMS step.
Range of values: integer value
Default value: -1
Required: no
* variance_encoded_in_target The flag that denotes if variance is encoded in target.
Range of values: {False, True}
Default value: False
Required: no
* keep_top_k Maximum number of bounding boxes per batch to be kept after NMS step.
Range of values: integer values
Default value: None
Required: yes
* code_type The type of coding method for bounding boxes.
Range of values: {'caffe.PriorBoxParameter.CENTER_SIZE',
'caffe.PriorBoxParameter.CORNER'}
Default value: 'caffe.PriorBoxParameter.CORNER'
Required: no
* share_location The flag that denotes if bounding boxes are shared among different
classes.
Range of values: {True, False}
Default value: True
Required: no
* nms_threshold The threshold to be used in the NMS stage.
Range of values: floating point value
Default value: None
Required: yes
* confidence_threshold Specifies the minimum confidence threshold for detection boxes to be
considered.
Range of values: floating point value
Default value: 0
Required: no
* clip_after_nms The flag that denotes whether to perform clip bounding boxes after
non-maximum suppression or not.
Range of values: {True, False}
Default value: False
Required: no
* clip_before_nms The flag that denotes whether to perform clip bounding boxes before
non-maximum suppression or not.
Range of values: {True, False}
Default value: False
Required: no
* decrease_label_id The flag that denotes how to perform NMS.
Range of values: False - perform NMS like in Caffe*.
True - perform NMS like in MxNet*.
Default value: False
Required: no
* normalized The flag that denotes whether input tensors with boxes are normalized.
Range of values: {True, False}
Default value: False
Required: no
* input_height The input image height.
Range of values: positive integer number
Default value: 1
Required: no
* input_width The input image width.
Range of values: positive integer number
Default value: 1
Required: no
* objectness_score The threshold to sort out confidence predictions.
Range of values: non-negative float number
Default value: 0
Required: no
Example of attribute dictionary:
@code{.py}
# just required ones
attrs = {
'keep_top_k': [1, 2, 3],
'nms_threshold': 0.645,
}
attrs = {
'keep_top_k': [1, 2, 3],
'nms_threshold': 0.645,
'normalized': True,
'clip_before_nms': True,
'input_height': [32],
'input_width': [32],
}
@endcode
Optional attributes which are absent from dictionary will be set with corresponding default.
"""
requirements = [
("background_label_id", False, np.integer, None),
("top_k", False, np.integer, None),
("variance_encoded_in_target", False, np.bool_, None),
("keep_top_k", True, np.integer, None),
("code_type", False, np.str_, None),
("share_location", False, np.bool_, None),
("nms_threshold", True, np.floating, None),
("confidence_threshold", False, np.floating, None),
("clip_after_nms", False, np.bool_, None),
("clip_before_nms", False, np.bool_, None),
("decrease_label_id", False, np.bool_, None),
("normalized", False, np.bool_, None),
("input_height", False, np.integer, is_positive_value),
("input_width", False, np.integer, is_positive_value),
("objectness_score", False, np.floating, is_non_negative_value),
]
check_valid_attributes("DetectionOutput", attrs, requirements)
inputs = [box_logits, class_preds, proposals]
if aux_class_preds is not None:
inputs.append(aux_class_preds)
if aux_box_preds is not None:
inputs.append(aux_box_preds)
inputs = as_nodes(*inputs)
return _get_node_factory_opset8().create("DetectionOutput", inputs, attrs)
def prior_box(layer_shape: Node,
image_shape: NodeInput,
attrs: dict,
name: Optional[str] = None) -> Node:
"""Generate prior boxes of specified sizes and aspect ratios across all dimensions.
@param layer_shape: Shape of layer for which prior boxes are computed.
@param image_shape: Shape of image to which prior boxes are scaled.
@param attrs: The dictionary containing key, value pairs for attributes.
@param name: Optional name for the output node.
@return Node representing prior box operation.
Available attributes are:
* min_size The minimum box size (in pixels).
Range of values: positive floating point numbers
Default value: []
Required: no
* max_size The maximum box size (in pixels).
Range of values: positive floating point numbers
Default value: []
Required: no
* aspect_ratio Aspect ratios of prior boxes.
Range of values: set of positive floating point numbers
Default value: []
Required: no
* flip The flag that denotes that each aspect_ratio is duplicated and flipped.
Range of values: {True, False}
Default value: False
Required: no
* clip The flag that denotes if each value in the output tensor should be clipped
to [0,1] interval.
Range of values: {True, False}
Default value: False
Required: no
* step The distance between box centers.
Range of values: floating point non-negative number
Default value: 0
Required: no
* offset This is a shift of box respectively to top left corner.
Range of values: floating point non-negative number
Default value: None
Required: yes
* variance The variance denotes a variance of adjusting bounding boxes. The attribute
could contain 0, 1 or 4 elements.
Range of values: floating point positive numbers
Default value: []
Required: no
* scale_all_sizes The flag that denotes type of inference.
Range of values: False - max_size is ignored
True - max_size is used
Default value: True
Required: no
* fixed_ratio This is an aspect ratio of a box.
Range of values: a list of positive floating-point numbers
Default value: None
Required: no
* fixed_size This is an initial box size (in pixels).
Range of values: a list of positive floating-point numbers
Default value: None
Required: no
* density This is the square root of the number of boxes of each type.
Range of values: a list of positive floating-point numbers
Default value: None
Required: no
* min_max_aspect_ratios_order The flag that denotes the order of output prior box.
Range of values: False - the output prior box is in order of
[min, aspect_ratios, max]
True - the output prior box is in order of
[min, max, aspect_ratios]
Default value: True
Required: no
Example of attribute dictionary:
@code{.py}
# just required ones
attrs = {
'offset': 85,
}
attrs = {
'offset': 85,
'flip': True,
'clip': True,
'fixed_size': [32, 64, 128]
}
@endcode
Optional attributes which are absent from dictionary will be set with corresponding default.
"""
requirements = [
("offset", True, np.floating, is_non_negative_value),
("min_size", False, np.floating, is_positive_value),
("max_size", False, np.floating, is_positive_value),
("aspect_ratio", False, np.floating, is_positive_value),
("flip", False, np.bool_, None),
("clip", False, np.bool_, None),
("step", False, np.floating, is_non_negative_value),
("variance", False, np.floating, is_positive_value),
("scale_all_sizes", False, np.bool_, None),
("fixed_ratio", False, np.floating, is_positive_value),
("fixed_size", False, np.floating, is_positive_value),
("density", False, np.floating, is_positive_value),
("min_max_aspect_ratios_order", False, np.bool_, None),
]
check_valid_attributes("PriorBox", attrs, requirements)
return _get_node_factory_opset8().create(
"PriorBox", [layer_shape, as_node(image_shape)], attrs)
def proposal(
class_probs: Node,
bbox_deltas: Node,
image_shape: NodeInput,
attrs: dict,
name: Optional[str] = None,
) -> Node:
"""Filter bounding boxes and outputs only those with the highest prediction confidence.
@param class_probs: 4D input floating point tensor with class prediction scores.
@param bbox_deltas: 4D input floating point tensor with corrected predictions of bounding boxes
@param image_shape: The 1D input tensor with 3 or 4 elements describing image shape.
@param attrs: The dictionary containing key, value pairs for attributes.
@param name: Optional name for the output node.
* base_size The size of the anchor to which scale and ratio attributes are applied.
Range of values: a positive unsigned integer number
Default value: None
Required: yes
* pre_nms_topn The number of bounding boxes before the NMS operation.
Range of values: a positive unsigned integer number
Default value: None
Required: yes
* post_nms_topn The number of bounding boxes after the NMS operation.
Range of values: a positive unsigned integer number
Default value: None
Required: yes
* nms_thresh The minimum value of the proposal to be taken into consideration.
Range of values: a positive floating-point number
Default value: None
Required: yes
* feat_stride The step size to slide over boxes (in pixels).
Range of values: a positive unsigned integer
Default value: None
Required: yes
* min_size The minimum size of box to be taken into consideration.
Range of values: a positive unsigned integer number
Default value: None
Required: yes
* ratio The ratios for anchor generation.
Range of values: a list of floating-point numbers
Default value: None
Required: yes
* scale The scales for anchor generation.
Range of values: a list of floating-point numbers
Default value: None
Required: yes
* clip_before_nms The flag that specifies whether to perform clip bounding boxes before
non-maximum suppression or not.
Range of values: True or False
Default value: True
Required: no
* clip_after_nms The flag that specifies whether to perform clip bounding boxes after
non-maximum suppression or not.
Range of values: True or False
Default value: False
Required: no
* normalize The flag that specifies whether to perform normalization of output boxes to
[0,1] interval or not.
Range of values: True or False
Default value: False
Required: no
* box_size_scale Specifies the scale factor applied to logits of box sizes before decoding.
Range of values: a positive floating-point number
Default value: 1.0
Required: no
* box_coordinate_scale Specifies the scale factor applied to logits of box coordinates
before decoding.
Range of values: a positive floating-point number
Default value: 1.0
Required: no
* framework Specifies how the box coordinates are calculated.
Range of values: "" (empty string) - calculate box coordinates like in Caffe*
tensorflow - calculate box coordinates like in the TensorFlow*
Object Detection API models
Default value: "" (empty string)
Required: no
Example of attribute dictionary:
@code{.py}
# just required ones
attrs = {
'base_size': 85,
'pre_nms_topn': 10,
'post_nms_topn': 20,
'nms_thresh': 0.34,
'feat_stride': 16,
'min_size': 32,
'ratio': [0.1, 1.5, 2.0, 2.5],
'scale': [2, 3, 3, 4],
}
@endcode
Optional attributes which are absent from dictionary will be set with corresponding default.
@return Node representing Proposal operation.
"""
requirements = [
("base_size", True, np.unsignedinteger, is_positive_value),
("pre_nms_topn", True, np.unsignedinteger, is_positive_value),
("post_nms_topn", True, np.unsignedinteger, is_positive_value),
("nms_thresh", True, np.floating, is_positive_value),
("feat_stride", True, np.unsignedinteger, is_positive_value),
("min_size", True, np.unsignedinteger, is_positive_value),
("ratio", True, np.floating, None),
("scale", True, np.floating, None),
("clip_before_nms", False, np.bool_, None),
("clip_after_nms", False, np.bool_, None),
("normalize", False, np.bool_, None),
("box_size_scale", False, np.floating, is_positive_value),
("box_coordinate_scale", False, np.floating, is_positive_value),
("framework", False, np.str_, None),
]
check_valid_attributes("Proposal", attrs, requirements)
return _get_node_factory_opset4().create(
"Proposal", [class_probs, bbox_deltas,
as_node(image_shape)], attrs)