def test_global_avgpool2d(change_ordering):
if not tf.test.gpu_device_name() and not change_ordering:
pytest.skip(
"Skip! Since tensorflow AvgPoolingOp op currently only supports the NHWC tensor format on the CPU"
)
model = LayerTest()
model.eval()
input_np = np.random.uniform(0, 1, (1, 3, 224, 224))
error = convert_and_test(model,
input_np,
verbose=False,
change_ordering=change_ordering)
def test_densenet(change_ordering):
if not tf.test.gpu_device_name() and not change_ordering:
pytest.skip(
"Skip! Since tensorflow Conv2D op currently only supports the NHWC tensor format on the CPU"
)
model = densenet121()
model.eval()
input_np = np.random.uniform(0, 1, (1, 3, 224, 224))
error = convert_and_test(model,
input_np,
verbose=False,
change_ordering=change_ordering)
def test_avgpool2d(change_ordering, kernel_size, padding, stride):
if not tf.test.gpu_device_name() and not change_ordering:
pytest.skip(
"Skip! Since tensorflow AvgPoolingOp op currently only supports the NHWC tensor format on the CPU"
)
if padding > kernel_size / 2:
# RuntimeError: invalid argument 2: pad should be smaller than half of kernel size,
# but got padW = 1, padH = 1, kW = 1,
pytest.skip("pad should be smaller than half of kernel size")
model = LayerTest(kernel_size=kernel_size, padding=padding, stride=stride)
model.eval()
input_np = np.random.uniform(0, 1, (1, 3, 224, 224))
error = convert_and_test(model,
input_np,
verbose=False,
change_ordering=change_ordering)
def func(change_ordering, kernel_size, padding, stride, bias, dilation,
groups):
if not tf.test.gpu_device_name() and not change_ordering:
pytest.skip(
"Skip! Since tensorflow Conv2D op currently only supports the NHWC tensor format on the CPU"
)
if stride > 1 and dilation > 1:
pytest.skip(
"strides > 1 not supported in conjunction with dilation_rate > 1")
model = LayerTest(groups * 3,
groups,
kernel_size=kernel_size,
padding=padding,
stride=stride,
bias=bias,
dilation=dilation,
groups=groups)
model.eval()
input_np = np.random.uniform(0, 1, (1, groups * 3, 224, 224))
error = convert_and_test(model,
input_np,
verbose=False,
change_ordering=change_ordering)
def test_f_sigmoid(change_ordering):
model = FSigmoid()
model.eval()
input_np = np.random.uniform(0, 1, (1, 3, 224, 224))
error = convert_and_test(model, input_np, verbose=False, change_ordering=change_ordering)
def test_convtranspose2d(change_ordering, kernel_size, padding, stride, bias):
outs = np.random.choice([1, 3, 7])
model = LayerTest(3, outs, kernel_size=kernel_size, padding=padding, stride=stride, bias=bias)
model.eval()
input_np = np.random.uniform(0, 1, (1, 3, 224, 224))
error = convert_and_test(model, input_np, verbose=False, change_ordering=change_ordering)
def test_layer_upsamle(change_ordering, mode, size, scale_factor):
model = LayerUpsample(mode=mode, size=size, scale_factor=scale_factor)
model.eval()
input_np = np.random.uniform(0, 1, (1, 3, 64, 64))
error = convert_and_test(model, input_np, verbose=False, change_ordering=change_ordering)
def test_f_interpole(change_ordering, mode, size, scale_factor):
model = FInterpolate(mode=mode, size=size, scale_factor=scale_factor)
model.eval()
input_np = np.random.uniform(0, 1, (1, 3, 64, 64))
error = convert_and_test(model, input_np, verbose=False, change_ordering=change_ordering)
def test_f_softmax(change_ordering, dim):
model = FSoftmax(dim)
model.eval()
input_np = np.random.uniform(0, 1, (1, 3, 224, 224))
error = convert_and_test(model, input_np, verbose=False, change_ordering=change_ordering)
def test_f_hardtanh(change_ordering):
model = LayerHardtanh()
model.eval()
input_np = np.random.uniform(0, 1, (1, 3, 224, 224))
error = convert_and_test(model, input_np, verbose=False, change_ordering=change_ordering)