| @@ -24,12 +24,12 @@ from mindspore.common.initializer import initializer | |||||
| def bias_init_zeros(shape): | def bias_init_zeros(shape): | ||||
| """Bias init method.""" | """Bias init method.""" | ||||
| return Tensor(np.array(np.zeros(shape).astype(np.float32)).astype(np.float16)) | |||||
| return Tensor(np.array(np.zeros(shape).astype(np.float32)), dtype=mstype.float32) | |||||
| def _conv(in_channels, out_channels, kernel_size=3, stride=1, padding=0, pad_mode='pad'): | def _conv(in_channels, out_channels, kernel_size=3, stride=1, padding=0, pad_mode='pad'): | ||||
| """Conv2D wrapper.""" | """Conv2D wrapper.""" | ||||
| shape = (out_channels, in_channels, kernel_size, kernel_size) | shape = (out_channels, in_channels, kernel_size, kernel_size) | ||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float16) | |||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32) | |||||
| shape_bias = (out_channels,) | shape_bias = (out_channels,) | ||||
| biass = bias_init_zeros(shape_bias) | biass = bias_init_zeros(shape_bias) | ||||
| return nn.Conv2d(in_channels, out_channels, | return nn.Conv2d(in_channels, out_channels, | ||||
| @@ -76,8 +76,10 @@ class FeatPyramidNeck(nn.Cell): | |||||
| self.fpn_convs_ = [] | self.fpn_convs_ = [] | ||||
| for _, channel in enumerate(in_channels): | for _, channel in enumerate(in_channels): | ||||
| l_conv = _conv(channel, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='valid') | |||||
| fpn_conv = _conv(out_channels, out_channels, kernel_size=3, stride=1, padding=0, pad_mode='same') | |||||
| l_conv = _conv(channel, out_channels, kernel_size=1, stride=1, | |||||
| padding=0, pad_mode='valid').to_float(mstype.float16) | |||||
| fpn_conv = _conv(out_channels, out_channels, kernel_size=3, stride=1, | |||||
| padding=0, pad_mode='same').to_float(mstype.float16) | |||||
| self.lateral_convs_list_.append(l_conv) | self.lateral_convs_list_.append(l_conv) | ||||
| self.fpn_convs_.append(fpn_conv) | self.fpn_convs_.append(fpn_conv) | ||||
| self.lateral_convs_list = nn.layer.CellList(self.lateral_convs_list_) | self.lateral_convs_list = nn.layer.CellList(self.lateral_convs_list_) | ||||
| @@ -26,8 +26,8 @@ class DenseNoTranpose(nn.Cell): | |||||
| """Dense method""" | """Dense method""" | ||||
| def __init__(self, input_channels, output_channels, weight_init): | def __init__(self, input_channels, output_channels, weight_init): | ||||
| super(DenseNoTranpose, self).__init__() | super(DenseNoTranpose, self).__init__() | ||||
| self.weight = Parameter(initializer(weight_init, [input_channels, output_channels], mstype.float16)) | |||||
| self.bias = Parameter(initializer("zeros", [output_channels], mstype.float16)) | |||||
| self.weight = Parameter(initializer(weight_init, [input_channels, output_channels], mstype.float32)) | |||||
| self.bias = Parameter(initializer("zeros", [output_channels], mstype.float32)) | |||||
| self.matmul = P.MatMul(transpose_b=False) | self.matmul = P.MatMul(transpose_b=False) | ||||
| self.bias_add = P.BiasAdd() | self.bias_add = P.BiasAdd() | ||||
| @@ -41,18 +41,18 @@ class FpnCls(nn.Cell): | |||||
| super(FpnCls, self).__init__() | super(FpnCls, self).__init__() | ||||
| representation_size = input_channels * pool_size * pool_size | representation_size = input_channels * pool_size * pool_size | ||||
| shape_0 = (output_channels, representation_size) | shape_0 = (output_channels, representation_size) | ||||
| weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=mstype.float16) | |||||
| weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=mstype.float32) | |||||
| shape_1 = (output_channels, output_channels) | shape_1 = (output_channels, output_channels) | ||||
| weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=mstype.float16) | |||||
| self.shared_fc_0 = DenseNoTranpose(representation_size, output_channels, weights_0) | |||||
| self.shared_fc_1 = DenseNoTranpose(output_channels, output_channels, weights_1) | |||||
| weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=mstype.float32) | |||||
| self.shared_fc_0 = DenseNoTranpose(representation_size, output_channels, weights_0).to_float(mstype.float16) | |||||
| self.shared_fc_1 = DenseNoTranpose(output_channels, output_channels, weights_1).to_float(mstype.float16) | |||||
| cls_weight = initializer('Normal', shape=[num_classes, output_channels][::-1], | cls_weight = initializer('Normal', shape=[num_classes, output_channels][::-1], | ||||
| dtype=mstype.float16) | |||||
| dtype=mstype.float32) | |||||
| reg_weight = initializer('Normal', shape=[num_classes * 4, output_channels][::-1], | reg_weight = initializer('Normal', shape=[num_classes * 4, output_channels][::-1], | ||||
| dtype=mstype.float16) | |||||
| self.cls_scores = DenseNoTranpose(output_channels, num_classes, cls_weight) | |||||
| self.reg_scores = DenseNoTranpose(output_channels, num_classes * 4, reg_weight) | |||||
| dtype=mstype.float32) | |||||
| self.cls_scores = DenseNoTranpose(output_channels, num_classes, cls_weight).to_float(mstype.float16) | |||||
| self.reg_scores = DenseNoTranpose(output_channels, num_classes * 4, reg_weight).to_float(mstype.float16) | |||||
| self.relu = P.ReLU() | self.relu = P.ReLU() | ||||
| self.flatten = P.Flatten() | self.flatten = P.Flatten() | ||||
| @@ -24,9 +24,9 @@ from mindspore.common.initializer import initializer | |||||
| def _conv(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='pad'): | def _conv(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='pad'): | ||||
| """Conv2D wrapper.""" | """Conv2D wrapper.""" | ||||
| shape = (out_channels, in_channels, kernel_size, kernel_size) | shape = (out_channels, in_channels, kernel_size, kernel_size) | ||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float16) | |||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32) | |||||
| shape_bias = (out_channels,) | shape_bias = (out_channels,) | ||||
| bias = Tensor(np.array(np.zeros(shape_bias)).astype(np.float16)) | |||||
| bias = Tensor(np.array(np.zeros(shape_bias)).astype(np.float32)) | |||||
| return nn.Conv2d(in_channels, out_channels, | return nn.Conv2d(in_channels, out_channels, | ||||
| kernel_size=kernel_size, stride=stride, padding=padding, | kernel_size=kernel_size, stride=stride, padding=padding, | ||||
| pad_mode=pad_mode, weight_init=weights, has_bias=True, bias_init=bias) | pad_mode=pad_mode, weight_init=weights, has_bias=True, bias_init=bias) | ||||
| @@ -34,9 +34,9 @@ def _conv(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mod | |||||
| def _convTanspose(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='pad'): | def _convTanspose(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='pad'): | ||||
| """ConvTranspose wrapper.""" | """ConvTranspose wrapper.""" | ||||
| shape = (out_channels, in_channels, kernel_size, kernel_size) | shape = (out_channels, in_channels, kernel_size, kernel_size) | ||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float16) | |||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32) | |||||
| shape_bias = (out_channels,) | shape_bias = (out_channels,) | ||||
| bias = Tensor(np.array(np.zeros(shape_bias)).astype(np.float16)) | |||||
| bias = Tensor(np.array(np.zeros(shape_bias)).astype(np.float32)) | |||||
| return nn.Conv2dTranspose(in_channels, out_channels, | return nn.Conv2dTranspose(in_channels, out_channels, | ||||
| kernel_size=kernel_size, stride=stride, padding=padding, | kernel_size=kernel_size, stride=stride, padding=padding, | ||||
| pad_mode=pad_mode, weight_init=weights, has_bias=True, bias_init=bias) | pad_mode=pad_mode, weight_init=weights, has_bias=True, bias_init=bias) | ||||
| @@ -45,21 +45,27 @@ class FpnMask(nn.Cell): | |||||
| """conv layers of mask head""" | """conv layers of mask head""" | ||||
| def __init__(self, input_channels, output_channels, num_classes): | def __init__(self, input_channels, output_channels, num_classes): | ||||
| super(FpnMask, self).__init__() | super(FpnMask, self).__init__() | ||||
| self.mask_conv1 = _conv(input_channels, output_channels, kernel_size=3, pad_mode="same") | |||||
| self.mask_conv1 = _conv(input_channels, output_channels, kernel_size=3, | |||||
| pad_mode="same").to_float(mstype.float16) | |||||
| self.mask_relu1 = P.ReLU() | self.mask_relu1 = P.ReLU() | ||||
| self.mask_conv2 = _conv(output_channels, output_channels, kernel_size=3, pad_mode="same") | |||||
| self.mask_conv2 = _conv(output_channels, output_channels, kernel_size=3, | |||||
| pad_mode="same").to_float(mstype.float16) | |||||
| self.mask_relu2 = P.ReLU() | self.mask_relu2 = P.ReLU() | ||||
| self.mask_conv3 = _conv(output_channels, output_channels, kernel_size=3, pad_mode="same") | |||||
| self.mask_conv3 = _conv(output_channels, output_channels, kernel_size=3, | |||||
| pad_mode="same").to_float(mstype.float16) | |||||
| self.mask_relu3 = P.ReLU() | self.mask_relu3 = P.ReLU() | ||||
| self.mask_conv4 = _conv(output_channels, output_channels, kernel_size=3, pad_mode="same") | |||||
| self.mask_conv4 = _conv(output_channels, output_channels, kernel_size=3, | |||||
| pad_mode="same").to_float(mstype.float16) | |||||
| self.mask_relu4 = P.ReLU() | self.mask_relu4 = P.ReLU() | ||||
| self.mask_deconv5 = _convTanspose(output_channels, output_channels, kernel_size=2, stride=2, pad_mode="valid") | |||||
| self.mask_deconv5 = _convTanspose(output_channels, output_channels, kernel_size=2, | |||||
| stride=2, pad_mode="valid").to_float(mstype.float16) | |||||
| self.mask_relu5 = P.ReLU() | self.mask_relu5 = P.ReLU() | ||||
| self.mask_conv6 = _conv(output_channels, num_classes, kernel_size=1, stride=1, pad_mode="valid") | |||||
| self.mask_conv6 = _conv(output_channels, num_classes, kernel_size=1, stride=1, | |||||
| pad_mode="valid").to_float(mstype.float16) | |||||
| def construct(self, x): | def construct(self, x): | ||||
| x = self.mask_conv1(x) | x = self.mask_conv1(x) | ||||
| @@ -164,23 +164,23 @@ class RPN(nn.Cell): | |||||
| shp_weight_conv = (feat_channels, in_channels, 3, 3) | shp_weight_conv = (feat_channels, in_channels, 3, 3) | ||||
| shp_bias_conv = (feat_channels,) | shp_bias_conv = (feat_channels,) | ||||
| weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=mstype.float16) | |||||
| bias_conv = initializer(0, shape=shp_bias_conv, dtype=mstype.float16) | |||||
| weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=mstype.float32) | |||||
| bias_conv = initializer(0, shape=shp_bias_conv, dtype=mstype.float32) | |||||
| shp_weight_cls = (num_anchors * cls_out_channels, feat_channels, 1, 1) | shp_weight_cls = (num_anchors * cls_out_channels, feat_channels, 1, 1) | ||||
| shp_bias_cls = (num_anchors * cls_out_channels,) | shp_bias_cls = (num_anchors * cls_out_channels,) | ||||
| weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=mstype.float16) | |||||
| bias_cls = initializer(0, shape=shp_bias_cls, dtype=mstype.float16) | |||||
| weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=mstype.float32) | |||||
| bias_cls = initializer(0, shape=shp_bias_cls, dtype=mstype.float32) | |||||
| shp_weight_reg = (num_anchors * 4, feat_channels, 1, 1) | shp_weight_reg = (num_anchors * 4, feat_channels, 1, 1) | ||||
| shp_bias_reg = (num_anchors * 4,) | shp_bias_reg = (num_anchors * 4,) | ||||
| weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=mstype.float16) | |||||
| bias_reg = initializer(0, shape=shp_bias_reg, dtype=mstype.float16) | |||||
| weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=mstype.float32) | |||||
| bias_reg = initializer(0, shape=shp_bias_reg, dtype=mstype.float32) | |||||
| for i in range(num_layers): | for i in range(num_layers): | ||||
| rpn_layer.append(RpnRegClsBlock(in_channels, feat_channels, num_anchors, cls_out_channels, \ | rpn_layer.append(RpnRegClsBlock(in_channels, feat_channels, num_anchors, cls_out_channels, \ | ||||
| weight_conv, bias_conv, weight_cls, \ | weight_conv, bias_conv, weight_cls, \ | ||||
| bias_cls, weight_reg, bias_reg)) | |||||
| bias_cls, weight_reg, bias_reg).to_float(mstype.float16)) | |||||
| for i in range(1, num_layers): | for i in range(1, num_layers): | ||||
| rpn_layer[i].rpn_conv.weight = rpn_layer[0].rpn_conv.weight | rpn_layer[i].rpn_conv.weight = rpn_layer[0].rpn_conv.weight | ||||
| @@ -24,12 +24,12 @@ from mindspore.common.initializer import initializer | |||||
| def bias_init_zeros(shape): | def bias_init_zeros(shape): | ||||
| """Bias init method.""" | """Bias init method.""" | ||||
| return Tensor(np.array(np.zeros(shape).astype(np.float32)).astype(np.float16)) | |||||
| return Tensor(np.array(np.zeros(shape).astype(np.float32)), dtype=mstype.float32) | |||||
| def _conv(in_channels, out_channels, kernel_size=3, stride=1, padding=0, pad_mode='pad'): | def _conv(in_channels, out_channels, kernel_size=3, stride=1, padding=0, pad_mode='pad'): | ||||
| """Conv2D wrapper.""" | """Conv2D wrapper.""" | ||||
| shape = (out_channels, in_channels, kernel_size, kernel_size) | shape = (out_channels, in_channels, kernel_size, kernel_size) | ||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float16).to_tensor() | |||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32).to_tensor() | |||||
| shape_bias = (out_channels,) | shape_bias = (out_channels,) | ||||
| biass = bias_init_zeros(shape_bias) | biass = bias_init_zeros(shape_bias) | ||||
| return nn.Conv2d(in_channels, out_channels, | return nn.Conv2d(in_channels, out_channels, | ||||
| @@ -76,8 +76,10 @@ class FeatPyramidNeck(nn.Cell): | |||||
| self.fpn_convs_ = [] | self.fpn_convs_ = [] | ||||
| for _, channel in enumerate(in_channels): | for _, channel in enumerate(in_channels): | ||||
| l_conv = _conv(channel, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='valid') | |||||
| fpn_conv = _conv(out_channels, out_channels, kernel_size=3, stride=1, padding=0, pad_mode='same') | |||||
| l_conv = _conv(channel, out_channels, kernel_size=1, stride=1, padding=0, | |||||
| pad_mode='valid').to_float(mstype.float16) | |||||
| fpn_conv = _conv(out_channels, out_channels, kernel_size=3, stride=1, padding=0, | |||||
| pad_mode='same').to_float(mstype.float16) | |||||
| self.lateral_convs_list_.append(l_conv) | self.lateral_convs_list_.append(l_conv) | ||||
| self.fpn_convs_.append(fpn_conv) | self.fpn_convs_.append(fpn_conv) | ||||
| self.lateral_convs_list = nn.layer.CellList(self.lateral_convs_list_) | self.lateral_convs_list = nn.layer.CellList(self.lateral_convs_list_) | ||||
| @@ -26,9 +26,9 @@ class DenseNoTranpose(nn.Cell): | |||||
| """Dense method""" | """Dense method""" | ||||
| def __init__(self, input_channels, output_channels, weight_init): | def __init__(self, input_channels, output_channels, weight_init): | ||||
| super(DenseNoTranpose, self).__init__() | super(DenseNoTranpose, self).__init__() | ||||
| self.weight = Parameter(initializer(weight_init, [input_channels, output_channels], mstype.float16), | |||||
| self.weight = Parameter(initializer(weight_init, [input_channels, output_channels], mstype.float32), | |||||
| name="weight") | name="weight") | ||||
| self.bias = Parameter(initializer("zeros", [output_channels], mstype.float16), name="bias") | |||||
| self.bias = Parameter(initializer("zeros", [output_channels], mstype.float32), name="bias") | |||||
| self.matmul = P.MatMul(transpose_b=False) | self.matmul = P.MatMul(transpose_b=False) | ||||
| self.bias_add = P.BiasAdd() | self.bias_add = P.BiasAdd() | ||||
| @@ -42,18 +42,18 @@ class FpnCls(nn.Cell): | |||||
| super(FpnCls, self).__init__() | super(FpnCls, self).__init__() | ||||
| representation_size = input_channels * pool_size * pool_size | representation_size = input_channels * pool_size * pool_size | ||||
| shape_0 = (output_channels, representation_size) | shape_0 = (output_channels, representation_size) | ||||
| weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=mstype.float16) | |||||
| weights_0 = initializer("XavierUniform", shape=shape_0[::-1], dtype=mstype.float32) | |||||
| shape_1 = (output_channels, output_channels) | shape_1 = (output_channels, output_channels) | ||||
| weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=mstype.float16) | |||||
| self.shared_fc_0 = DenseNoTranpose(representation_size, output_channels, weights_0) | |||||
| self.shared_fc_1 = DenseNoTranpose(output_channels, output_channels, weights_1) | |||||
| weights_1 = initializer("XavierUniform", shape=shape_1[::-1], dtype=mstype.float32) | |||||
| self.shared_fc_0 = DenseNoTranpose(representation_size, output_channels, weights_0).to_float(mstype.float16) | |||||
| self.shared_fc_1 = DenseNoTranpose(output_channels, output_channels, weights_1).to_float(mstype.float16) | |||||
| cls_weight = initializer('Normal', shape=[num_classes, output_channels][::-1], | cls_weight = initializer('Normal', shape=[num_classes, output_channels][::-1], | ||||
| dtype=mstype.float16) | |||||
| dtype=mstype.float32) | |||||
| reg_weight = initializer('Normal', shape=[num_classes * 4, output_channels][::-1], | reg_weight = initializer('Normal', shape=[num_classes * 4, output_channels][::-1], | ||||
| dtype=mstype.float16) | |||||
| self.cls_scores = DenseNoTranpose(output_channels, num_classes, cls_weight) | |||||
| self.reg_scores = DenseNoTranpose(output_channels, num_classes * 4, reg_weight) | |||||
| dtype=mstype.float32) | |||||
| self.cls_scores = DenseNoTranpose(output_channels, num_classes, cls_weight).to_float(mstype.float16) | |||||
| self.reg_scores = DenseNoTranpose(output_channels, num_classes * 4, reg_weight).to_float(mstype.float16) | |||||
| self.relu = P.ReLU() | self.relu = P.ReLU() | ||||
| self.flatten = P.Flatten() | self.flatten = P.Flatten() | ||||
| @@ -24,9 +24,9 @@ from mindspore.common.initializer import initializer | |||||
| def _conv(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='pad'): | def _conv(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='pad'): | ||||
| """Conv2D wrapper.""" | """Conv2D wrapper.""" | ||||
| shape = (out_channels, in_channels, kernel_size, kernel_size) | shape = (out_channels, in_channels, kernel_size, kernel_size) | ||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float16) | |||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32) | |||||
| shape_bias = (out_channels,) | shape_bias = (out_channels,) | ||||
| bias = Tensor(np.array(np.zeros(shape_bias)).astype(np.float16)) | |||||
| bias = Tensor(np.array(np.zeros(shape_bias)).astype(np.float32)) | |||||
| return nn.Conv2d(in_channels, out_channels, | return nn.Conv2d(in_channels, out_channels, | ||||
| kernel_size=kernel_size, stride=stride, padding=padding, | kernel_size=kernel_size, stride=stride, padding=padding, | ||||
| pad_mode=pad_mode, weight_init=weights, has_bias=True, bias_init=bias) | pad_mode=pad_mode, weight_init=weights, has_bias=True, bias_init=bias) | ||||
| @@ -34,9 +34,9 @@ def _conv(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mod | |||||
| def _convTanspose(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='pad'): | def _convTanspose(in_channels, out_channels, kernel_size=1, stride=1, padding=0, pad_mode='pad'): | ||||
| """ConvTranspose wrapper.""" | """ConvTranspose wrapper.""" | ||||
| shape = (out_channels, in_channels, kernel_size, kernel_size) | shape = (out_channels, in_channels, kernel_size, kernel_size) | ||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float16) | |||||
| weights = initializer("XavierUniform", shape=shape, dtype=mstype.float32) | |||||
| shape_bias = (out_channels,) | shape_bias = (out_channels,) | ||||
| bias = Tensor(np.array(np.zeros(shape_bias)).astype(np.float16)) | |||||
| bias = Tensor(np.array(np.zeros(shape_bias)).astype(np.float32)) | |||||
| return nn.Conv2dTranspose(in_channels, out_channels, | return nn.Conv2dTranspose(in_channels, out_channels, | ||||
| kernel_size=kernel_size, stride=stride, padding=padding, | kernel_size=kernel_size, stride=stride, padding=padding, | ||||
| pad_mode=pad_mode, weight_init=weights, has_bias=True, bias_init=bias) | pad_mode=pad_mode, weight_init=weights, has_bias=True, bias_init=bias) | ||||
| @@ -45,21 +45,27 @@ class FpnMask(nn.Cell): | |||||
| """conv layers of mask head""" | """conv layers of mask head""" | ||||
| def __init__(self, input_channels, output_channels, num_classes): | def __init__(self, input_channels, output_channels, num_classes): | ||||
| super(FpnMask, self).__init__() | super(FpnMask, self).__init__() | ||||
| self.mask_conv1 = _conv(input_channels, output_channels, kernel_size=3, pad_mode="same") | |||||
| self.mask_conv1 = _conv(input_channels, output_channels, kernel_size=3, | |||||
| pad_mode="same").to_float(mstype.float16) | |||||
| self.mask_relu1 = P.ReLU() | self.mask_relu1 = P.ReLU() | ||||
| self.mask_conv2 = _conv(output_channels, output_channels, kernel_size=3, pad_mode="same") | |||||
| self.mask_conv2 = _conv(output_channels, output_channels, kernel_size=3, | |||||
| pad_mode="same").to_float(mstype.float16) | |||||
| self.mask_relu2 = P.ReLU() | self.mask_relu2 = P.ReLU() | ||||
| self.mask_conv3 = _conv(output_channels, output_channels, kernel_size=3, pad_mode="same") | |||||
| self.mask_conv3 = _conv(output_channels, output_channels, kernel_size=3, | |||||
| pad_mode="same").to_float(mstype.float16) | |||||
| self.mask_relu3 = P.ReLU() | self.mask_relu3 = P.ReLU() | ||||
| self.mask_conv4 = _conv(output_channels, output_channels, kernel_size=3, pad_mode="same") | |||||
| self.mask_conv4 = _conv(output_channels, output_channels, kernel_size=3, | |||||
| pad_mode="same").to_float(mstype.float16) | |||||
| self.mask_relu4 = P.ReLU() | self.mask_relu4 = P.ReLU() | ||||
| self.mask_deconv5 = _convTanspose(output_channels, output_channels, kernel_size=2, stride=2, pad_mode="valid") | |||||
| self.mask_deconv5 = _convTanspose(output_channels, output_channels, kernel_size=2, | |||||
| stride=2, pad_mode="valid").to_float(mstype.float16) | |||||
| self.mask_relu5 = P.ReLU() | self.mask_relu5 = P.ReLU() | ||||
| self.mask_conv6 = _conv(output_channels, num_classes, kernel_size=1, stride=1, pad_mode="valid") | |||||
| self.mask_conv6 = _conv(output_channels, num_classes, kernel_size=1, stride=1, | |||||
| pad_mode="valid").to_float(mstype.float16) | |||||
| def construct(self, x): | def construct(self, x): | ||||
| x = self.mask_conv1(x) | x = self.mask_conv1(x) | ||||
| @@ -164,23 +164,23 @@ class RPN(nn.Cell): | |||||
| shp_weight_conv = (feat_channels, in_channels, 3, 3) | shp_weight_conv = (feat_channels, in_channels, 3, 3) | ||||
| shp_bias_conv = (feat_channels,) | shp_bias_conv = (feat_channels,) | ||||
| weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=mstype.float16) | |||||
| bias_conv = initializer(0, shape=shp_bias_conv, dtype=mstype.float16) | |||||
| weight_conv = initializer('Normal', shape=shp_weight_conv, dtype=mstype.float32) | |||||
| bias_conv = initializer(0, shape=shp_bias_conv, dtype=mstype.float32) | |||||
| shp_weight_cls = (num_anchors * cls_out_channels, feat_channels, 1, 1) | shp_weight_cls = (num_anchors * cls_out_channels, feat_channels, 1, 1) | ||||
| shp_bias_cls = (num_anchors * cls_out_channels,) | shp_bias_cls = (num_anchors * cls_out_channels,) | ||||
| weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=mstype.float16) | |||||
| bias_cls = initializer(0, shape=shp_bias_cls, dtype=mstype.float16) | |||||
| weight_cls = initializer('Normal', shape=shp_weight_cls, dtype=mstype.float32) | |||||
| bias_cls = initializer(0, shape=shp_bias_cls, dtype=mstype.float32) | |||||
| shp_weight_reg = (num_anchors * 4, feat_channels, 1, 1) | shp_weight_reg = (num_anchors * 4, feat_channels, 1, 1) | ||||
| shp_bias_reg = (num_anchors * 4,) | shp_bias_reg = (num_anchors * 4,) | ||||
| weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=mstype.float16) | |||||
| bias_reg = initializer(0, shape=shp_bias_reg, dtype=mstype.float16) | |||||
| weight_reg = initializer('Normal', shape=shp_weight_reg, dtype=mstype.float32) | |||||
| bias_reg = initializer(0, shape=shp_bias_reg, dtype=mstype.float32) | |||||
| for i in range(num_layers): | for i in range(num_layers): | ||||
| rpn_layer.append(RpnRegClsBlock(in_channels, feat_channels, num_anchors, cls_out_channels, \ | rpn_layer.append(RpnRegClsBlock(in_channels, feat_channels, num_anchors, cls_out_channels, \ | ||||
| weight_conv, bias_conv, weight_cls, \ | weight_conv, bias_conv, weight_cls, \ | ||||
| bias_cls, weight_reg, bias_reg)) | |||||
| bias_cls, weight_reg, bias_reg).to_float(mstype.float16)) | |||||
| for i in range(1, num_layers): | for i in range(1, num_layers): | ||||
| rpn_layer[i].rpn_conv.weight = rpn_layer[0].rpn_conv.weight | rpn_layer[i].rpn_conv.weight = rpn_layer[0].rpn_conv.weight | ||||