def __init__(self, channel=1, w=0.25):
super().__init__()
self.norm = P.L2Normalize(axis=1)
self.prelu = P.PReLU()
self.w = Parameter(initializer(w, [
channel,
]), name='w')
def __init__(self,
embbeding_size=128,
classnum=270762,
s=32,
a=1.0,
m=0.3,
b=0.2):
super(CombineMarginFCFp16, self).__init__()
weight_shape = [classnum, embbeding_size]
weight_init = initializer(me_init.ReidXavierUniform(), weight_shape)
self.weight = Parameter(weight_init, name='weight')
self.m = m
self.s = s
self.a = a
self.b = b
self.m_const = Tensor(self.m, dtype=mstype.float16)
self.a_const = Tensor(self.a, dtype=mstype.float16)
self.b_const = Tensor(self.b, dtype=mstype.float16)
self.s_const = Tensor(self.s, dtype=mstype.float16)
self.m_const_zero = Tensor(0, dtype=mstype.float16)
self.a_const_one = Tensor(1, dtype=mstype.float16)
self.normalize = P.L2Normalize(axis=1)
self.fc = P.MatMul(transpose_b=True)
self.onehot = P.OneHot()
self.transpose = P.Transpose()
self.acos = P.ACos()
self.cos = P.Cos()
self.cast = P.Cast()
self.on_value = Tensor(1.0, mstype.float32)
self.off_value = Tensor(0.0, mstype.float32)
def __init__(self, channel=1, w=0.25, strategy1=None, strategy2=None):
super().__init__()
self.norm = P.L2Normalize().set_strategy(strategy1)
self.prelu = P.PReLU().set_strategy(strategy2)
self.w = Parameter(initializer(w, [
channel,
]), name='w')
def __init__(self,
block,
layer_nums,
in_channels,
out_channels,
strides,
low_dims,
training_mode=True,
use_MLP=False):
super(ResNet, self).__init__()
if not len(layer_nums) == len(in_channels) == len(out_channels) == 4:
raise ValueError("the length of layer_num, in_channels, out_channels list must be 4!")
self.use_MLP = use_MLP
self.training_mode = training_mode
self.concat = P.Concat()
self.split = P.Split(0, 3)
self.l2norm = P.L2Normalize(axis=1)
self.conv1 = _conv3x3(3, 64, stride=1)
self.bn1 = _bn(64, training_mode)
self.relu = nn.ReLU()
self.layer1 = self._make_layer(block,
layer_nums[0],
in_channel=in_channels[0],
out_channel=out_channels[0],
stride=strides[0])
self.layer2 = self._make_layer(block,
layer_nums[1],
in_channel=in_channels[1],
out_channel=out_channels[1],
stride=strides[1])
self.layer3 = self._make_layer(block,
layer_nums[2],
in_channel=in_channels[2],
out_channel=out_channels[2],
stride=strides[2])
self.layer4 = self._make_layer(block,
layer_nums[3],
in_channel=in_channels[3],
out_channel=out_channels[3],
stride=strides[3])
self.mean = P.ReduceMean(keep_dims=True)
self.flatten = nn.Flatten()
self.end_point = _fc(block.expansion * 512, low_dims)
self.mlp_layer1 = _fc(block.expansion * 512, block.expansion * 512)
self.mlp_layer2 = _fc(block.expansion * 512, low_dims)
def __init__(self,
in_channels,
out_channels,
weight_init='normal',
bias_init='zeros',
has_bias=True):
super(CommonHeadLastFN, self).__init__()
weight_shape = [out_channels, in_channels]
self.weight = Parameter(initializer(weight_init, weight_shape),
requires_grad=True,
name='weight')
self.x_norm = P.L2Normalize(axis=1)
self.w_norm = P.L2Normalize(axis=1)
self.fc = P.MatMul(transpose_a=False, transpose_b=True)
self.multiplier = Parameter(Tensor(np.ones([1]), mstype.float32),
requires_grad=True,
name='multiplier')
self.has_bias = has_bias
if self.has_bias:
bias_shape = [out_channels]
self.bias_add = P.BiasAdd()
self.bias = Parameter(initializer(bias_init, bias_shape),
requires_grad=True,
name='bias')
def __init__(self, in_channel, out_channel, axis, input_shape, mul_size,
test_size, prelu_size, transpose_b, matmul_size, num_class):
super().__init__()
mul_np = np.full(mul_size, 0.5, dtype=np.float32)
self.mul_weight = Parameter(Tensor(mul_np), name="mul_weight")
bias_np = np.full((12, ), 7.1, dtype=np.float32)
self.bias = Parameter(Tensor(bias_np), name="bias")
prelu_np = np.full(prelu_size, 0.8, dtype=np.float32)
self.prelu_weight = Parameter(Tensor(prelu_np), name="prelu_weight")
matmul_np = np.full(matmul_size, 1.1, dtype=np.float32)
self.matmul_weight = Parameter(Tensor(matmul_np), name="matmul_weight")
self.mul = P.Mul()
self.conv = Conv2d(in_channels=in_channel,
out_channels=out_channel,
kernel_size=5,
has_bias=True,
weight_init='ones',
bias_init='ones',
pad_mode='valid')
self.scalar = 0.5
self.parameter = Parameter(initializer(0.5,
test_size,
dtype=mstype.float32),
name='parameter')
self.tensor = Tensor(np.full(test_size, 0.05, dtype=np.float32))
self.softmax = Softmax(axis=axis)
self.relu = ReLU()
self.relu.relu.add_prim_attr("primitive_target", "CPU")
self.reshape = P.Reshape()
self.input_shape = input_shape
self.equal = P.Equal()
self.cast = P.Cast()
self.concat = P.Concat(axis=1)
self.reduce_sum = P.ReduceSum()
self.bias_add = P.BiasAdd()
self.cos = P.Cos()
self.prelu = P.PReLU()
self.matmul = P.MatMul(transpose_b=transpose_b)
self.l2norm = P.L2Normalize(axis=(1 - axis))
self.tensoradd = P.TensorAdd()
self.strided_slice = P.StridedSlice()
self.dense = Dense(in_channels=6,
out_channels=num_class,
weight_init='ones',
bias_init='ones',
has_bias=True)
def __init__(self, num_layers=36, feature_dim=128, shape=(96, 64)):
super(SphereNet_float32, self).__init__()
assert num_layers in [12, 20, 36, 64], 'SphereNet num_layers should be 12, 20 or 64'
if num_layers == 12:
layers = [1, 1, 1, 1]
filter_list = [3, 16, 32, 64, 128]
fc_size = 128 * 6 * 4
elif num_layers == 20:
layers = [1, 2, 4, 1]
filter_list = [3, 64, 128, 256, 512]
fc_size = 512 * 6 * 4
elif num_layers == 36:
layers = [2, 4, 4, 2]
filter_list = [3, 32, 64, 128, 256]
fc_size = 256 * 6 * 4
elif num_layers == 64:
layers = [3, 7, 16, 3]
filter_list = [3, 64, 128, 256, 512]
fc_size = 512 * 6 * 4
else:
raise ValueError('sphere' + str(num_layers) + " IS NOT SUPPORTED! (sphere20 or sphere64)")
self.shape = P.Shape()
self.reshape = P.Reshape()
block = BaseBlock
self.layer1 = MakeLayer(block, filter_list[0], filter_list[1], layers[0], stride=2)
self.layer2 = MakeLayer(block, filter_list[1], filter_list[2], layers[1], stride=2)
self.layer3 = MakeLayer(block, filter_list[2], filter_list[3], layers[2], stride=2)
self.layer4 = MakeLayer(block, filter_list[3], filter_list[4], layers[3], stride=2)
self.fc = fc_with_initialize(fc_size, feature_dim)
self.last_bn = nn.BatchNorm1d(feature_dim, momentum=0.9).add_flags_recursive(fp32=True)
self.last_bn_sub = nn.BatchNorm2d(feature_dim, momentum=0.9).add_flags_recursive(fp32=True)
self.cast = P.Cast()
self.l2norm = P.L2Normalize(axis=1)
for _, cell in self.cells_and_names():
if isinstance(cell, (nn.Conv2d, nn.Dense)):
if cell.bias is not None:
cell.weight.set_data(initializer(me_init.ReidKaimingUniform(a=math.sqrt(5), mode='fan_out'),
cell.weight.shape))
cell.bias.set_data(initializer('zeros', cell.bias.shape))
else:
cell.weight.set_data(initializer(me_init.ReidXavierUniform(), cell.weight.shape))
self.device_target = context.get_context('device_target')
def __init__(self,
temperature=0.07,
contrast_mode='all',
base_temperature=0.07):
super(SupConLoss, self).__init__()
self.temperature = temperature
self.contrast_mode = contrast_mode
self.base_temperature = base_temperature
self.normalize = P.L2Normalize(axis=2)
self.eye = P.Eye()
self.unbind = P.Unstack(axis=1)
self.cat = P.Concat(axis=0)
self.matmul = P.MatMul()
self.div = P.Div()
self.transpose = P.Transpose()
self.maxes = P.ArgMaxWithValue(axis=1, keep_dims=True)
self.tile = P.Tile()
self.scatter = P.ScatterNd()
self.oneslike = P.OnesLike()
self.exp = P.Exp()
self.sum = P.ReduceSum(keep_dims=True)
self.log = P.Log()
self.reshape = P.Reshape()
self.mean = P.ReduceMean()
def __init__(self, args):
super(CombineMarginFC, self).__init__()
weight_shape = [args.num_classes, args.emb_size]
weight_init = initializer(me_init.ReidXavierUniform(), weight_shape)
self.weight = Parameter(weight_init, name='weight')
self.m = args.margin_m
self.s = args.margin_s
self.a = args.margin_a
self.b = args.margin_b
self.m_const = Tensor(self.m, dtype=mstype.float16)
self.a_const = Tensor(self.a, dtype=mstype.float16)
self.b_const = Tensor(self.b, dtype=mstype.float16)
self.s_const = Tensor(self.s, dtype=mstype.float16)
self.m_const_zero = Tensor(0, dtype=mstype.float16)
self.a_const_one = Tensor(1, dtype=mstype.float16)
self.normalize = P.L2Normalize(axis=1)
self.fc = P.MatMul(transpose_b=True)
self.onehot = P.OneHot()
self.transpose = P.Transpose()
self.acos = P.ACos()
self.cos = P.Cos()
self.cast = P.Cast()
self.on_value = Tensor(1.0, mstype.float32)
self.off_value = Tensor(0.0, mstype.float32)
def __init__(self):
super(L2NormalizeNet, self).__init__()
self.l2_normalize = P.L2Normalize()
def __init__(self,
in_channels,
out_channels,
weight_init='normal',
bias_init='zeros',
damping=0.03,
loss_scale=1,
frequency=278,
batch_size=32.0,
has_bias=True,
activation=None):
super(Dense_ThorNoBN, self).__init__()
self.batch_size = batch_size
self.in_channels = check_int_positive(in_channels)
self.out_channels = check_int_positive(out_channels)
self.has_bias = check_bool(has_bias)
self.thor = True
if isinstance(weight_init, Tensor):
if weight_init.dim() != 2 or weight_init.shape()[0] != out_channels or \
weight_init.shape()[1] != in_channels:
raise ValueError("weight_init shape error")
self.weight = Parameter(initializer(weight_init,
[out_channels, in_channels]),
name="weight")
if self.has_bias:
if isinstance(bias_init, Tensor):
if bias_init.dim() != 1 or bias_init.shape(
)[0] != out_channels:
raise ValueError("bias_init shape error")
self.bias = Parameter(initializer(bias_init, [out_channels]),
name="bias")
self.matmul = P.MatMul(transpose_b=True)
self.bias_add = P.BiasAdd()
self.activation = get_activation(activation)
self.activation_flag = self.activation is not None
self.matrix_A_inv = Parameter(Tensor(
np.zeros([128, 128, 16, 16]).astype(np.float16)),
name='matrix_A_inv',
requires_grad=False)
self.matrix_G_inv = Parameter(Tensor(
np.zeros([63, 63, 16, 16]).astype(np.float16)),
name="matrix_G_inv",
requires_grad=False)
self.fake_G = Tensor(np.zeros([63, 63, 16, 16]).astype(np.float16))
self.matmul = P.MatMul(transpose_b=True)
self.cube_matmul = P.CusMatMulCube(transpose_a=True)
self.matrix_combine = P.CusMatrixCombine()
self.cholesky = P.CusCholeskyTrsm()
self.shape = P.Shape()
self.reshape = P.Reshape()
self.transpose = P.Transpose()
self.cov_step = Parameter(initializer(0, [1], mstype.int32),
name="cov_step",
requires_grad=False)
self.mul = P.Mul()
self.cast = P.Cast()
self.damping = Tensor(damping)
self.loss_scale = Tensor(1 / loss_scale, mstype.float16)
self.vector_matmul = P.CusBatchMatMul()
self.pad = P.Pad(((0, 24), (0, 24)))
self.pad1 = P.Pad(((0, 8), (0, 8)))
self.slice = P.Slice()
self.gather = P.GatherV2()
self.assignadd = P.AssignAdd()
self.freq = Tensor(frequency, mstype.int32)
self.axis = 0
self.A_inv_max = Parameter(initializer(0, [1], mstype.float32),
name="A_inv_max",
requires_grad=False)
self.G_inv_max = Parameter(initializer(0, [1], mstype.float32),
name="G_inv_max",
requires_grad=False)
self.fused_abs_max1 = P.CusFusedAbsMax1([1000, 1000])
self.fused_abs_max2 = P.CusFusedAbsMax1()
self.log = P.Log()
self.exp = P.Exp()
self.dampingA = Tensor(np.identity(2048), mstype.float32)
self.dampingG = Tensor(np.identity(1024), mstype.float32)
self.add = P.TensorAdd()
self.sqrt = P.Sqrt()
self.getG = P.InsertGradientOf(self.save_gradient)
self.fake_G_inv_max = Tensor(np.zeros([
1,
]).astype(np.float32))
self.norm = P.L2Normalize(axis=1)
self.multiplier = Parameter(Tensor(np.ones([1]), dtype=mstype.float16),
requires_grad=True,
name='multiplier')
def __init__(self, axis=0, epsilon=1e-4):
super(Net, self).__init__()
self.norm = P.L2Normalize(axis=axis, epsilon=epsilon)
def __init__(self, axis=0, epsilon=1e-4, strategy0=None, strategy1=None):
super(L2normalize, self).__init__()
self.add = P.TensorAdd(strategy=strategy0)
self.l2norm = P.L2Normalize(axis, epsilon, strategy1)
def __init__(self, strategy1, strategy2, strategy3):
super().__init__()
self.norm1 = P.L2Normalize(axis=0).shard(strategy1)
self.norm2 = P.L2Normalize(axis=0).shard(strategy1)
self.mul1 = P.Mul().shard(strategy2)
self.mul2 = P.Mul().shard(strategy3)
def __init__(self):
super().__init__()
self.norm1 = P.L2Normalize()
self.norm2 = P.L2Normalize()
self.mul1 = P.Mul()
self.mul2 = P.Mul()
def __init__(self, args, strategy):
super(SemiAutoOneHotNet, self).__init__()
self.a = args.a
self.b = args.b
self.c = args.c
self.d = args.d
self.e = args.e
self.cast = P.Cast()
self.cast.set_strategy(strategy=strategy.twod_strategy)
self.cast1 = P.Cast()
self.cast1.set_strategy(strategy=strategy.twod_strategy)
self.cast2 = P.Cast()
self.cast2.set_strategy(strategy=strategy.twod_strategy)
self.cast3 = P.Cast()
self.cast3.set_strategy(strategy=strategy.scalar_strategy)
self.cast4 = P.Cast()
self.cast4.set_strategy(strategy=strategy.scalar_strategy)
self.a_const = Tensor(self.a, dtype=mstype.float32)
self.b_const = Tensor(self.b, dtype=mstype.float32)
self.c_const = Tensor(self.c, dtype=mstype.float32)
self.d_const = Tensor(self.d, dtype=mstype.float32)
self.e_const = Tensor(self.e, dtype=mstype.float32)
self.m_const_zero = Tensor(0, dtype=mstype.float32)
self.a_const_one = Tensor(1, dtype=mstype.float32)
self.onehot = P.OneHot()
self.onehot.set_strategy(strategy=strategy.onehot_strategy)
self.exp = P.Exp()
self.exp.set_strategy(strategy=strategy.twod_strategy)
self.exp2 = P.Exp()
self.exp2.set_strategy(strategy=strategy.twod_strategy)
self.exp3 = P.Exp()
self.exp3.set_strategy(strategy=strategy.twod_strategy)
self.mul_const = P.Mul()
self.mul_const.set_strategy(strategy=strategy.scalar_twod_strategy)
self.mul_const2 = P.TensorAdd()
self.mul_const2.set_strategy(strategy=strategy.scalar_twod_strategy)
self.mul_const3 = P.Sub()
self.mul_const3.set_strategy(strategy=strategy.twod_scalar_strategy)
self.mul_const4 = P.Sub()
self.mul_const4.set_strategy(strategy=strategy.scalar_twod_strategy)
self.mul_const5 = P.Mul()
self.mul_const5.set_strategy(strategy=strategy.twod_scalar_strategy)
self.mul = P.Mul()
self.mul.set_strategy(strategy=strategy.twod_twod_strategy)
self.mul2 = P.Mul()
self.mul2.set_strategy(strategy=strategy.twod_twod_strategy)
self.mul3 = P.TensorAdd()
self.mul3.set_strategy(strategy=strategy.twod_twod_strategy)
self.mul4 = P.Sub()
self.mul4.set_strategy(strategy=strategy.twod_twodbc_strategy)
self.mul5 = P.RealDiv()
self.mul5.set_strategy(strategy=strategy.twod_twodbc_strategy)
self.mul6 = P.Mul()
self.mul6.set_strategy(strategy=strategy.twod_twod_strategy)
self.mul7 = P.Mul()
self.mul7.set_strategy(strategy=strategy.twod_scalar_strategy)
self.mul8 = P.RealDiv()
self.mul8.set_strategy(strategy=strategy.scalar_scalar_strategy)
self.mul9 = P.TensorAdd()
self.mul9.set_strategy(strategy=strategy.twod_scalar_strategy)
self.reduce_max = P.ReduceMax(keep_dims=True)
self.reduce_max.set_strategy(strategy=strategy.twod_strategy)
self.reduce_sum = P.ReduceSum(keep_dims=False)
self.reduce_sum.set_strategy(strategy=strategy.twod_strategy)
self.reduce_sum_2 = P.ReduceSum(keep_dims=False)
self.reduce_sum_2.set_strategy(strategy=strategy.twod_strategy)
self.reduce_sum_3 = P.ReduceSum(keep_dims=False)
self.reduce_sum_3.set_strategy(strategy=strategy.oned_strategy)
self.reshape = P.Reshape()
self.log = P.Log()
self.log.set_strategy(strategy=strategy.twod_strategy)
self.on_value = Tensor(1.0, mstype.float32)
self.off_value = Tensor(0.0, mstype.float32)
self.normalize = P.L2Normalize(axis=1)
self.normalize.set_strategy(strategy=strategy.twod_strategy_m)
self.normalize2 = P.L2Normalize(axis=1)
self.normalize2.set_strategy(strategy=strategy.twod_strategy_m)
self.fc = P.MatMul(transpose_b=True)
self.fc.set_strategy(strategy=strategy.twodbc_twod_strategy)
weight_shape = [args.num_classes, args.emb_size]
weight_np = np.zeros(weight_shape, np.float32)
self.weight = Parameter(Tensor(weight_np),
name='model_parallel_weight')
('Sigmoid', {
'block': P.Sigmoid(),
'desc_inputs': [[4, 128, 1024]],
'desc_bprop': [[4, 128, 1024]]}),
('Softmax', {
'block': P.Softmax(),
'desc_inputs': [[1, 16]],
'desc_bprop': [[1, 16]],
'skip': ['backward']}), # check backward error
('Softmax', {
'block': P.Softmax(axis=(0, 1)),
'desc_inputs': [[1, 16]],
'desc_bprop': [[1, 16]],
'skip': ['backward']}),
('L2Normalize', {
'block': P.L2Normalize(),
'desc_inputs': [[4, 128, 1024]],
'desc_bprop': [[4, 128, 1024]]}),
('ReLU', {
'block': P.ReLU(),
'desc_inputs': [[64, 64, 112, 112]],
'desc_bprop': [[64, 64, 112, 112]]}),
('SeqConvBnRelu', {
'block': SeqConvBnRelu(3, 64),
'desc_inputs': [[64, 3, 112, 112]],
'desc_bprop': [[64, 64, 112, 112]]}),
('PReluCell', {
'block': nn.PReLU(1, [np.float32(0.25)]),
'desc_inputs': [[128, 64, 112, 112]],
'desc_bprop': [[128, 64, 112, 112]]}),
('PRelu', {
