def test_proposal(int_dtype, fp_dtype):
attributes = {
"attrs.base_size": int_dtype(1),
"attrs.pre_nms_topn": int_dtype(20),
"attrs.post_nms_topn": int_dtype(64),
"attrs.nms_thresh": fp_dtype(0.34),
"attrs.feat_stride": int_dtype(16),
"attrs.min_size": int_dtype(32),
"attrs.ratio": np.array([0.1, 1.5, 2.0, 2.5], dtype=fp_dtype),
"attrs.scale": np.array([2, 3, 3, 4], dtype=fp_dtype),
}
batch_size = 7
class_probs = ng.parameter([batch_size, 12, 34, 62], fp_dtype,
"class_probs")
class_logits = ng.parameter([batch_size, 24, 34, 62], fp_dtype,
"class_logits")
image_shape = ng.parameter([3], fp_dtype, "image_shape")
node = ng.proposal(class_probs, class_logits, image_shape, attributes)
assert node.get_type_name() == "Proposal"
assert node.get_output_size() == 1
assert list(node.get_output_shape(0)) == [
batch_size * attributes["attrs.post_nms_topn"], 5
]
def test_proposal_props():
float_dtype = np.float32
batch_size = 1
post_nms_topn = 20
probs = ng.parameter(Shape([batch_size, 8, 255, 255]),
dtype=float_dtype,
name="probs")
deltas = ng.parameter(Shape([batch_size, 16, 255, 255]),
dtype=float_dtype,
name="bbox_deltas")
im_info = ng.parameter(Shape([4]), dtype=float_dtype, name="im_info")
attrs = {
"base_size": np.uint32(85),
"pre_nms_topn": np.uint32(10),
"post_nms_topn": np.uint32(post_nms_topn),
"nms_thresh": np.float32(0.34),
"feat_stride": np.uint32(16),
"min_size": np.uint32(32),
"ratio": np.array([0.1, 1.5, 2.0, 2.5], dtype=np.float32),
"scale": np.array([2, 3, 3, 4], dtype=np.float32),
}
node = ng.proposal(probs, deltas, im_info, attrs)
assert node.get_type_name() == "Proposal"
assert node.get_output_size() == 2
assert list(node.get_output_shape(0)) == [batch_size * post_nms_topn, 5]
assert list(node.get_output_shape(1)) == [batch_size * post_nms_topn]
assert node.get_output_element_type(0) == Type.f32
assert node.get_output_element_type(1) == Type.f32
def test_dynamic_set_attribute_value(int_dtype, fp_dtype):
attributes = {
"base_size": int_dtype(1),
"pre_nms_topn": int_dtype(20),
"post_nms_topn": int_dtype(64),
"nms_thresh": fp_dtype(0.34),
"feat_stride": int_dtype(16),
"min_size": int_dtype(32),
"ratio": np.array([0.1, 1.5, 2.0, 2.5], dtype=fp_dtype),
"scale": np.array([2, 3, 3, 4], dtype=fp_dtype),
}
batch_size = 7
class_probs = ng.parameter([batch_size, 12, 34, 62], fp_dtype,
"class_probs")
bbox_deltas = ng.parameter([batch_size, 24, 34, 62], fp_dtype,
"bbox_deltas")
image_shape = ng.parameter([3], fp_dtype, "image_shape")
node = ng.proposal(class_probs, bbox_deltas, image_shape, attributes)
node.set_base_size(int_dtype(15))
node.set_pre_nms_topn(int_dtype(7))
node.set_post_nms_topn(int_dtype(33))
node.set_nms_thresh(fp_dtype(1.55))
node.set_feat_stride(int_dtype(8))
node.set_min_size(int_dtype(123))
node.set_ratio(np.array([1.1, 2.5, 3.0, 4.5], dtype=fp_dtype))
node.set_scale(np.array([2.1, 3.2, 3.3, 4.4], dtype=fp_dtype))
node.set_clip_before_nms(True)
node.set_clip_after_nms(True)
node.set_normalize(True)
node.set_box_size_scale(fp_dtype(1.34))
node.set_box_coordinate_scale(fp_dtype(0.88))
node.set_framework("OpenVINO")
assert node.get_base_size() == int_dtype(15)
assert node.get_pre_nms_topn() == int_dtype(7)
assert node.get_post_nms_topn() == int_dtype(33)
assert np.isclose(node.get_nms_thresh(), fp_dtype(1.55))
assert node.get_feat_stride() == int_dtype(8)
assert node.get_min_size() == int_dtype(123)
assert np.allclose(node.get_ratio(),
np.array([1.1, 2.5, 3.0, 4.5], dtype=fp_dtype))
assert np.allclose(node.get_scale(),
np.array([2.1, 3.2, 3.3, 4.4], dtype=fp_dtype))
assert node.get_clip_before_nms()
assert node.get_clip_after_nms()
assert node.get_normalize()
assert np.isclose(node.get_box_size_scale(), fp_dtype(1.34))
assert np.isclose(node.get_box_coordinate_scale(), fp_dtype(0.88))
assert node.get_framework() == "OpenVINO"
def _proposal_node():
attributes = {
"base_size": np.uint16(1),
"pre_nms_topn": np.uint16(20),
"post_nms_topn": np.uint16(64),
"nms_thresh": np.float64(0.34),
"feat_stride": np.uint16(16),
"min_size": np.uint16(32),
"ratio": np.array([0.1, 1.5, 2.0, 2.5], dtype=np.float64),
"scale": np.array([2, 3, 3, 4], dtype=np.float64),
}
batch_size = 7
class_probs = ng.parameter([batch_size, 12, 34, 62], np.float64, "class_probs")
bbox_deltas = ng.parameter([batch_size, 24, 34, 62], np.float64, "bbox_deltas")
image_shape = ng.parameter([3], np.float64, "image_shape")
return ng.proposal(class_probs, bbox_deltas, image_shape, attributes)
def _proposal_node():
attributes = {
"attrs.base_size": np.uint16(1),
"attrs.pre_nms_topn": np.uint16(20),
"attrs.post_nms_topn": np.uint16(64),
"attrs.nms_thresh": np.float64(0.34),
"attrs.feat_stride": np.uint16(16),
"attrs.min_size": np.uint16(32),
"attrs.ratio": np.array([0.1, 1.5, 2.0, 2.5], dtype=np.float64),
"attrs.scale": np.array([2, 3, 3, 4], dtype=np.float64),
}
batch_size = 7
class_probs = ng.parameter([batch_size, 12, 34, 62], np.float64, "class_probs")
class_logits = ng.parameter([batch_size, 24, 34, 62], np.float64, "class_logits")
image_shape = ng.parameter([3], np.float64, "image_shape")
return ng.proposal(class_probs, class_logits, image_shape, attributes)
