def create_fast_rcnn_predictor(conv_out, rois, fc_layers, cfg):
# RCNN
roi_out = roipooling(conv_out,
rois,
cntk.MAX_POOLING,
(cfg["MODEL"].ROI_DIM, cfg["MODEL"].ROI_DIM),
spatial_scale=1 / 16.0)
fc_out = fc_layers(roi_out)
# prediction head
W_pred = parameter(shape=(4096, cfg["DATA"].NUM_CLASSES),
init=normal(scale=0.01),
name="cls_score.W")
b_pred = parameter(shape=cfg["DATA"].NUM_CLASSES,
init=0,
name="cls_score.b")
cls_score = plus(times(fc_out, W_pred), b_pred, name='cls_score')
# regression head
W_regr = parameter(shape=(4096, cfg["DATA"].NUM_CLASSES * 4),
init=normal(scale=0.001),
name="bbox_regr.W")
b_regr = parameter(shape=cfg["DATA"].NUM_CLASSES * 4,
init=0,
name="bbox_regr.b")
bbox_pred = plus(times(fc_out, W_regr), b_regr, name='bbox_regr')
return cls_score, bbox_pred
def create_fast_rcnn_predictor(conv_out, rois, fc_layers):
# RCNN
roi_out = roipooling(conv_out,
rois,
cntk.MAX_POOLING, (roi_dim, roi_dim),
spatial_scale=1 / 16.0)
print("roi_out shape:{}".format(roi_out.shape))
fc_out = fc_layers(roi_out)
# prediction head
fea_map_dim = globalvars['fea_map_dim']
W_pred = parameter(shape=(fea_map_dim, globalvars['num_classes']),
init=normal(scale=0.01),
name="cls_score.W")
b_pred = parameter(shape=globalvars['num_classes'],
init=0,
name="cls_score.b")
cls_score = plus(times(fc_out, W_pred), b_pred, name='cls_score')
# regression head
W_regr = parameter(shape=(fea_map_dim, globalvars['num_classes'] * 4),
init=normal(scale=0.001),
name="bbox_regr.W")
b_regr = parameter(shape=globalvars['num_classes'] * 4,
init=0,
name="bbox_regr.b")
bbox_pred = plus(times(fc_out, W_regr), b_regr, name='bbox_regr')
return cls_score, bbox_pred
def test_MaxRoiPool(tmpdir):
input_map = [[
[1., 2., 3.], # (1, 3, 3) input operand (conv feature map)
[4., 5., 6.],
[7., 8., 9.]
]]
input_rois = [[1, 1, 2, 2]]
conv_input = np.asarray(input_map, dtype=np.float32)
roi_input = np.asarray(input_rois, dtype=np.float32)
a = C.input_variable(shape=conv_input.shape,
dtype=np.float32,
needs_gradient=True,
name='a')
b = C.input_variable(shape=roi_input.shape,
dtype=np.float32,
needs_gradient=False,
name='b')
# adding batch and sequence axis
conv_input.shape = (1, ) + conv_input.shape
roi_input.shape = (1, ) + roi_input.shape
model = C.roipooling(a, b, C.MAX_POOLING, (3, 3), 1.)
verify_two_input(model, conv_input, roi_input, tmpdir, 'MaxRoiPool_1')
def test_MaxRoiPool(tmpdir, dtype):
pytest.skip(
'MaxRoiPool is failing with ONNX shape inference (input rois). RuntimeError: [ShapeInferenceError] RoIs tensor must have 2 dimensions'
)
with C.default_options(dtype=dtype):
input_map = [[
[1., 2., 3.], # (1, 3, 3) input operand (conv feature map)
[4., 5., 6.],
[7., 8., 9.]
]]
input_rois = [[1, 1, 2, 2]]
conv_input = np.asarray(input_map, dtype=dtype)
roi_input = np.asarray(input_rois, dtype=dtype)
a = C.input_variable(shape=conv_input.shape,
dtype=dtype,
needs_gradient=True,
name='a')
b = C.input_variable(shape=roi_input.shape,
dtype=dtype,
needs_gradient=False,
name='b')
# adding batch and sequence axis
conv_input.shape = (1, ) + conv_input.shape
roi_input.shape = (1, ) + roi_input.shape
model = C.roipooling(a, b, C.MAX_POOLING, (3, 3), 1.)
verify_two_input(model, conv_input, roi_input, tmpdir, 'MaxRoiPool_1')
def test_MaxRoiPool(tmpdir, dtype):
pytest.skip('MaxRoiPool is failing with ONNX shape inference (input rois). RuntimeError: [ShapeInferenceError] RoIs tensor must have 2 dimensions')
with C.default_options(dtype = dtype):
input_map = [[[1., 2., 3.], # (1, 3, 3) input operand (conv feature map)
[4., 5., 6.],
[7., 8., 9.]]]
input_rois = [[1, 1, 2, 2]]
conv_input = np.asarray(input_map, dtype=dtype)
roi_input = np.asarray(input_rois, dtype=dtype)
a = C.input_variable(shape=conv_input.shape,
dtype=dtype,
needs_gradient=True,
name='a')
b = C.input_variable(shape=roi_input.shape,
dtype=dtype,
needs_gradient=False,
name='b')
# adding batch and sequence axis
conv_input.shape = (1,) + conv_input.shape
roi_input.shape = (1,) + roi_input.shape
model = C.roipooling(a, b, C.MAX_POOLING, (3,3), 1.)
verify_two_input(model, conv_input, roi_input, tmpdir, 'MaxRoiPool_1')
def test_MaxRoiPool(tmpdir):
input_map = [[[1., 2., 3.], # (1, 3, 3) input operand (conv feature map)
[4., 5., 6.],
[7., 8., 9.]]]
input_rois = [[1, 1, 2, 2]]
conv_input = np.asarray(input_map, dtype=np.float32)
roi_input = np.asarray(input_rois, dtype=np.float32)
a = C.input_variable(shape=conv_input.shape,
dtype=np.float32,
needs_gradient=True,
name='a')
b = C.input_variable(shape=roi_input.shape,
dtype=np.float32,
needs_gradient=False,
name='b')
# adding batch and sequence axis
conv_input.shape = (1,) + conv_input.shape
roi_input.shape = (1,) + roi_input.shape
model = C.roipooling(a, b, C.MAX_POOLING, (3,3), 1.)
verify_two_input(model, conv_input, roi_input, tmpdir, 'MaxRoiPool_1')
def create_fast_rcnn_predictor(conv_out, rois, fc_layers, cfg):
# RCNN
roi_out = roipooling(conv_out, rois, cntk.MAX_POOLING, (cfg["MODEL"].ROI_DIM, cfg["MODEL"].ROI_DIM), spatial_scale=1/16.0)
fc_out = fc_layers(roi_out)
# prediction head
W_pred = parameter(shape=(4096, cfg["DATA"].NUM_CLASSES), init=normal(scale=0.01), name="cls_score.W")
b_pred = parameter(shape=cfg["DATA"].NUM_CLASSES, init=0, name="cls_score.b")
cls_score = plus(times(fc_out, W_pred), b_pred, name='cls_score')
# regression head
W_regr = parameter(shape=(4096, cfg["DATA"].NUM_CLASSES*4), init=normal(scale=0.001), name="bbox_regr.W")
b_regr = parameter(shape=cfg["DATA"].NUM_CLASSES*4, init=0, name="bbox_regr.b")
bbox_pred = plus(times(fc_out, W_regr), b_regr, name='bbox_regr')
return cls_score, bbox_pred
def create_fast_rcnn_predictor(conv_out, rois, fc_layers):
# RCNN
roi_out = roipooling(conv_out, rois, cntk.MAX_POOLING, (roi_dim, roi_dim), spatial_scale=1/16.0)
fc_out = fc_layers(roi_out)
# prediction head
W_pred = parameter(shape=(4096, globalvars['num_classes']), init=normal(scale=0.01), name="cls_score.W")
b_pred = parameter(shape=globalvars['num_classes'], init=0, name="cls_score.b")
cls_score = plus(times(fc_out, W_pred), b_pred, name='cls_score')
# regression head
W_regr = parameter(shape=(4096, globalvars['num_classes']*4), init=normal(scale=0.001), name="bbox_regr.W")
b_regr = parameter(shape=globalvars['num_classes']*4, init=0, name="bbox_regr.b")
bbox_pred = plus(times(fc_out, W_regr), b_regr, name='bbox_regr')
return cls_score, bbox_pred
def test_op_roipooling(input_map, input_rois, expected_fwd, expected_bkwd,
device_id, precision):
dt = PRECISION_TO_TYPE[precision]
# AA == as numpy array
conv_input = AA(input_map, dtype=dt)
roi_input = AA(input_rois, dtype=dt)
exp_fwd_value = AA(expected_fwd, dtype=dt)
exp_bkwd_value = AA(expected_bkwd, dtype=dt)
# adding batch, sequence and roi axis
exp_fwd_value.shape = (1, 1, 1) + exp_fwd_value.shape
exp_bkwd_value.shape = (1, 1) + exp_bkwd_value.shape
# I == define cntk input variables
a = I(shape=conv_input.shape,
dtype=sanitize_dtype_cntk(precision),
needs_gradient=True,
name='a')
b = I(shape=roi_input.shape,
dtype=sanitize_dtype_cntk(precision),
needs_gradient=False,
name='b')
# adding batch and sequence axis
conv_input.shape = (1, 1) + conv_input.shape
roi_input.shape = (1, 1) + roi_input.shape
from cntk import roipooling
input_op = roipooling(a, b, (3, 3))
forward_input = {a: conv_input, b: roi_input}
expected_backward = {a: exp_bkwd_value}
unittest_helper(input_op,
forward_input,
exp_fwd_value,
expected_backward,
device_id=device_id,
precision=precision)
def test_op_maxroipooling(input_map, input_rois, expected_fwd, expected_bkwd, device_id, precision):
dt = PRECISION_TO_TYPE[precision]
# AA == as numpy array
conv_input = AA(input_map, dtype=dt)
roi_input = AA(input_rois, dtype=dt)
exp_fwd_value = AA(expected_fwd, dtype=dt)
exp_bkwd_value = AA(expected_bkwd, dtype=dt)
# adding batch, sequence and roi axis
exp_fwd_value.shape = (1,1) + exp_fwd_value.shape
exp_bkwd_value.shape = (1,) + exp_bkwd_value.shape
# I == define cntk input variables
a = C.input_variable(shape=conv_input.shape,
dtype=sanitize_dtype_cntk(precision),
needs_gradient=True,
name='a')
b = C.input_variable(shape=roi_input.shape,
dtype=sanitize_dtype_cntk(precision),
needs_gradient=False,
name='b')
# adding batch and sequence axis
conv_input.shape = (1,) + conv_input.shape
roi_input.shape = (1,) + roi_input.shape
from cntk import roipooling
input_op = roipooling(a, b, C.MAX_POOLING, (3,3), 1.)
forward_input = {a: conv_input, b: roi_input}
expected_backward = {a: exp_bkwd_value}
unittest_helper(input_op,
forward_input, exp_fwd_value, expected_backward,
device_id=device_id, precision=precision)