def get_symbol(num_classes, version, **kwargs):
"""Get symbol of SqueezeNet
Parameters
----------
num_classes: int
The number of classification results
version : str, optional
"1.0" or "1.1" of SqueezeNet
"""
assert version == '1.1', ("Unsupported SqueezeNet version {version}:"
"1.1 expected".format(version=version))
net = sym.Variable("data")
net = sym.conv2d(net, channels=64, kernel_size=(3, 3), strides=(2, 2))
net = sym.relu(net)
net = sym.max_pool2d(net, pool_size=(3, 3), strides=(2, 2))
net = _make_fire(net, 16, 64, 64)
net = _make_fire(net, 16, 64, 64)
net = sym.max_pool2d(net, pool_size=(3, 3), strides=(2, 2))
net = _make_fire(net, 32, 128, 128)
net = _make_fire(net, 32, 128, 128)
net = sym.max_pool2d(net, pool_size=(3, 3), strides=(2, 2))
net = _make_fire(net, 48, 192, 192)
net = _make_fire(net, 48, 192, 192)
net = _make_fire(net, 64, 256, 256)
net = _make_fire(net, 64, 256, 256)
net = sym.dropout(net, rate=0.5)
net = sym.conv2d(net, channels=num_classes, kernel_size=(1, 1))
net = sym.relu(net)
net = sym.global_avg_pool2d(net)
return sym.softmax(net, axis=1)
def test_injective_conv2d():
channels = 16
data = sym.Variable(name="data")
pool = sym.global_avg_pool2d(data=data)
weight = sym.reshape(pool, shape=[1, channels, 1, 1])
residual = sym.conv2d(data=data, kernel_size=(3,3), channels=channels, padding=(1, 1),
layout="NCHW", kernel_layout="OIHW", use_bias=False, name="conv")
net = weight * data + residual
size = 56
dtype="float32"
dshape = (1, channels, size, size)
kshape = (channels, channels, 3, 3)
oshape = dshape
shape_dict = {"data": dshape}
for target, ctx in ctx_list():
graph, lib, _ = nnvm.compiler.build(net, target, shape_dict)
# data, global_avg_pool, conv weight, conv op, fused elemwise add
assert graph.index.num_nodes == 5
data = tvm.nd.array(np.random.uniform(size=dshape).astype(dtype))
kernel = tvm.nd.array(np.random.uniform(size=kshape).astype(dtype))
m = graph_runtime.create(graph, lib, ctx)
m.run(data=data, conv_weight=kernel)
# get output
out = m.get_output(0, tvm.nd.empty(oshape, dtype))
residual = topi.testing.conv2d_nchw_python(
data.asnumpy(), kernel.asnumpy(), (1,1), 'SAME')
weight = np.mean(data.asnumpy(), axis=(2, 3))
c_np = weight[:, :, np.newaxis, np.newaxis] * data.asnumpy() + residual
tvm.testing.assert_allclose(out.asnumpy(), c_np, rtol=1e-5)
def test_global_avg_pool2d():
x = sym.Variable("x")
y = sym.global_avg_pool2d(x, name="y")
dtype = "float32"
dshape = (1, 1024, 7, 7)
oshape = (1, 1024, 1, 1)
shape_dict = {"x": dshape}
for target, ctx in ctx_list():
graph, lib, _ = nnvm.compiler.build(y, target, shape_dict)
m = graph_runtime.create(graph, lib, ctx)
data = tvm.nd.array(np.random.uniform(size=dshape).astype(dtype))
m.run(x=data)
out = m.get_output(0, tvm.nd.empty(oshape, dtype))
b_np = np.mean(data.asnumpy(), axis=(2,3), keepdims=True)
tvm.testing.assert_allclose(out.asnumpy(), b_np, rtol=1e-5)
def test_global_avg_pool2d():
x = sym.Variable("x")
y = sym.global_avg_pool2d(x, name="y")
dtype = "float32"
dshape = (1, 1024, 7, 7)
oshape = (1, 1024, 1, 1)
shape_dict = {"x": dshape}
for target, ctx in ctx_list():
graph, lib, _ = nnvm.compiler.build(y, target, shape_dict)
m = graph_runtime.create(graph, lib, ctx)
data = tvm.nd.array(np.random.uniform(size=dshape).astype(dtype))
m.run(x=data)
out = m.get_output(0, tvm.nd.empty(oshape, dtype))
b_np = np.mean(data.asnumpy(), axis=(2, 3), keepdims=True)
np.testing.assert_allclose(out.asnumpy(), b_np, rtol=1e-5)
def test_injective_conv2d():
channels = 16
data = sym.Variable(name="data")
pool = sym.global_avg_pool2d(data=data)
weight = sym.reshape(pool, shape=[1, channels, 1, 1])
residual = sym.conv2d(data=data,
kernel_size=(3, 3),
channels=channels,
padding=(1, 1),
layout="NCHW",
kernel_layout="OIHW",
use_bias=False,
name="conv")
net = weight * data + residual
size = 56
dtype = "float32"
dshape = (1, channels, size, size)
kshape = (channels, channels, 3, 3)
oshape = dshape
shape_dict = {"data": dshape}
for target, ctx in ctx_list():
graph, lib, _ = nnvm.compiler.build(net, target, shape_dict)
# data, global_avg_pool, conv weight, conv op, fused elemwise add
assert graph.index.num_nodes == 5
data = tvm.nd.array(np.random.uniform(size=dshape).astype(dtype))
kernel = tvm.nd.array(np.random.uniform(size=kshape).astype(dtype))
m = graph_runtime.create(graph, lib, ctx)
m.run(data=data, conv_weight=kernel)
# get output
out = m.get_output(0, tvm.nd.empty(oshape, dtype))
residual = topi.testing.conv2d_nchw_python(data.asnumpy(),
kernel.asnumpy(), (1, 1),
'SAME')
weight = np.mean(data.asnumpy(), axis=(2, 3))
c_np = weight[:, :, np.newaxis, np.newaxis] * data.asnumpy() + residual
np.testing.assert_allclose(out.asnumpy(), c_np, rtol=1e-5)
def alter_global_avg_pool2d_layout(attrs, inputs, tinfos):
new_attrs = {k: attrs[k] for k in attrs.keys()}
new_attrs["layout"] = "NCHW16c"
return sym.global_avg_pool2d(inputs[0], **new_attrs)