项目
数据集
课程
比赛
模型库
活动
更多
论坛
访问飞桨官网
首页 Paddle框架 帖子详情
谁来用低阶api写个vgg瞅瞅呗
收藏
快速回复
Paddle框架 问答深度学习模型训练 556 20
谁来用低阶api写个vgg瞅瞅呗
收藏
快速回复
Paddle框架 问答深度学习模型训练 556 20

谁来用低阶api写个vgg瞅瞅呗

看源码看不懂。

就洗个vgg19

 

https://aistudio.baidu.com/aistudio/projectdetail/1540993

0
收藏
回复
全部评论(20)
时间顺序
AIStudio810259
#2 回复于2021-02

搞好了。

0
回复
AIStudio810259
#3 回复于2021-02
AIStudio810259 #2
搞好了。

还是不行，醉了。。。

0
回复
AIStudio810258
#4 回复于2021-02

vgg就是直接码ConvBN

0
回复
AIStudio810258
#5 回复于2021-02

然后该pool就pool，没啥啊

0
回复
AIStudio810258
#6 回复于2021-02

不行用paddleclass套件里的vgg试试

 

0
回复
AIStudio810258
#7 回复于2021-02

import paddle
from paddle import ParamAttr
import paddle.nn as nn
import paddle.nn.functional as F
from paddle.nn import Conv2D, BatchNorm, Linear, Dropout
from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D

__all__ = ["VGG11", "VGG13", "VGG16", "VGG19"]


class ConvBlock(nn.Layer):
    def __init__(self, input_channels, output_channels, groups, name=None):
        super(ConvBlock, self).__init__()

        self.groups = groups
        self._conv_1 = Conv2D(
            in_channels=input_channels,
            out_channels=output_channels,
            kernel_size=3,
            stride=1,
            padding=1,
            weight_attr=ParamAttr(name=name + "1_weights"),
            bias_attr=False)
        if groups == 2 or groups == 3 or groups == 4:
            self._conv_2 = Conv2D(
                in_channels=output_channels,
                out_channels=output_channels,
                kernel_size=3,
                stride=1,
                padding=1,
                weight_attr=ParamAttr(name=name + "2_weights"),
                bias_attr=False)
        if groups == 3 or groups == 4:
            self._conv_3 = Conv2D(
                in_channels=output_channels,
                out_channels=output_channels,
                kernel_size=3,
                stride=1,
                padding=1,
                weight_attr=ParamAttr(name=name + "3_weights"),
                bias_attr=False)
        if groups == 4:
            self._conv_4 = Conv2D(
                in_channels=output_channels,
                out_channels=output_channels,
                kernel_size=3,
                stride=1,
                padding=1,
                weight_attr=ParamAttr(name=name + "4_weights"),
                bias_attr=False)

        self._pool = MaxPool2D(kernel_size=2, stride=2, padding=0)

    def forward(self, inputs):
        x = self._conv_1(inputs)
        x = F.relu(x)
        if self.groups == 2 or self.groups == 3 or self.groups == 4:
            x = self._conv_2(x)
            x = F.relu(x)
        if self.groups == 3 or self.groups == 4:
            x = self._conv_3(x)
            x = F.relu(x)
        if self.groups == 4:
            x = self._conv_4(x)
            x = F.relu(x)
        x = self._pool(x)
        return x


class VGGNet(nn.Layer):
    def __init__(self, layers=11, class_dim=1000):
        super(VGGNet, self).__init__()

        self.layers = layers
        self.vgg_configure = {
            11: [1, 1, 2, 2, 2],
            13: [2, 2, 2, 2, 2],
            16: [2, 2, 3, 3, 3],
            19: [2, 2, 4, 4, 4]
        }
        assert self.layers in self.vgg_configure.keys(), \
            "supported layers are {} but input layer is {}".format(
                vgg_configure.keys(), layers)
        self.groups = self.vgg_configure[self.layers]

        self._conv_block_1 = ConvBlock(3, 64, self.groups[0], name="conv1_")
        self._conv_block_2 = ConvBlock(64, 128, self.groups[1], name="conv2_")
        self._conv_block_3 = ConvBlock(128, 256, self.groups[2], name="conv3_")
        self._conv_block_4 = ConvBlock(256, 512, self.groups[3], name="conv4_")
        self._conv_block_5 = ConvBlock(512, 512, self.groups[4], name="conv5_")

        self._drop = Dropout(p=0.5, mode="downscale_in_infer")
        self._fc1 = Linear(
            7 * 7 * 512,
            4096,
            weight_attr=ParamAttr(name="fc6_weights"),
            bias_attr=ParamAttr(name="fc6_offset"))
        self._fc2 = Linear(
            4096,
            4096,
            weight_attr=ParamAttr(name="fc7_weights"),
            bias_attr=ParamAttr(name="fc7_offset"))
        self._out = Linear(
            4096,
            class_dim,
            weight_attr=ParamAttr(name="fc8_weights"),
            bias_attr=ParamAttr(name="fc8_offset"))

    def forward(self, inputs):
        x = self._conv_block_1(inputs)
        x = self._conv_block_2(x)
        x = self._conv_block_3(x)
        x = self._conv_block_4(x)
        x = self._conv_block_5(x)

        x = paddle.reshape(x, [0, -1])
        x = self._fc1(x)
        x = F.relu(x)
        x = self._drop(x)
        x = self._fc2(x)
        x = F.relu(x)
        x = self._drop(x)
        x = self._out(x)
        return x


def VGG11(**args):
    model = VGGNet(layers=11, **args)
    return model


def VGG13(**args):
    model = VGGNet(layers=13, **args)
    return model


def VGG16(**args):
    model = VGGNet(layers=16, **args)
    return model


def VGG19(**args):
    model = VGGNet(layers=19, **args)
    return model

 

0
回复
AIStudio810258
#8 回复于2021-02
AIStudio810258 #7
import paddle from paddle import ParamAttr import paddle.nn as nn import paddle.nn.functional as F from paddle.nn import Conv2D, BatchNorm, Linear, Dropout from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D __all__ = ["VGG11", "VGG13", "VGG16", "VGG19"] class ConvBlock(nn.Layer):     def __init__(self, input_channels, output_channels, groups, name=None):         super(ConvBlock, self).__init__()         self.groups = groups         self._conv_1 = Conv2D(             in_channels=input_channels,             out_channels=output_channels,             kernel_size=3,             stride=1,             padding=1,             weight_attr=ParamAttr(name=name + "1_weights"),             bias_attr=False)         if groups == 2 or groups == 3 or groups == 4:             self._conv_2 = Conv2D(                 in_channels=output_channels,                 out_channels=output_channels,                 kernel_size=3,                 stride=1,                 padding=1,                 weight_attr=ParamAttr(name=name + "2_weights"),                 bias_attr=False)         if groups == 3 or groups == 4:             self._conv_3 = Conv2D(                 in_channels=output_channels,                 out_channels=output_channels,                 kernel_size=3,                 stride=1,                 padding=1,                 weight_attr=ParamAttr(name=name + "3_weights"),                 bias_attr=False)         if groups == 4:             self._conv_4 = Conv2D(                 in_channels=output_channels,                 out_channels=output_channels,                 kernel_size=3,                 stride=1,                 padding=1,                 weight_attr=ParamAttr(name=name + "4_weights"),                 bias_attr=False)         self._pool = MaxPool2D(kernel_size=2, stride=2, padding=0)     def forward(self, inputs):         x = self._conv_1(inputs)         x = F.relu(x)         if self.groups == 2 or self.groups == 3 or self.groups == 4:             x = self._conv_2(x)             x = F.relu(x)         if self.groups == 3 or self.groups == 4:             x = self._conv_3(x)             x = F.relu(x)         if self.groups == 4:             x = self._conv_4(x)             x = F.relu(x)         x = self._pool(x)         return x class VGGNet(nn.Layer):     def __init__(self, layers=11, class_dim=1000):         super(VGGNet, self).__init__()         self.layers = layers         self.vgg_configure = {             11: [1, 1, 2, 2, 2],             13: [2, 2, 2, 2, 2],             16: [2, 2, 3, 3, 3],             19: [2, 2, 4, 4, 4]         }         assert self.layers in self.vgg_configure.keys(), \             "supported layers are {} but input layer is {}".format(                 vgg_configure.keys(), layers)         self.groups = self.vgg_configure[self.layers]         self._conv_block_1 = ConvBlock(3, 64, self.groups[0], name="conv1_")         self._conv_block_2 = ConvBlock(64, 128, self.groups[1], name="conv2_")         self._conv_block_3 = ConvBlock(128, 256, self.groups[2], name="conv3_")         self._conv_block_4 = ConvBlock(256, 512, self.groups[3], name="conv4_")         self._conv_block_5 = ConvBlock(512, 512, self.groups[4], name="conv5_")         self._drop = Dropout(p=0.5, mode="downscale_in_infer")         self._fc1 = Linear(             7 * 7 * 512,             4096,             weight_attr=ParamAttr(name="fc6_weights"),             bias_attr=ParamAttr(name="fc6_offset"))         self._fc2 = Linear(             4096,             4096,             weight_attr=ParamAttr(name="fc7_weights"),             bias_attr=ParamAttr(name="fc7_offset"))         self._out = Linear(             4096,             class_dim,             weight_attr=ParamAttr(name="fc8_weights"),             bias_attr=ParamAttr(name="fc8_offset"))     def forward(self, inputs):         x = self._conv_block_1(inputs)         x = self._conv_block_2(x)         x = self._conv_block_3(x)         x = self._conv_block_4(x)         x = self._conv_block_5(x)         x = paddle.reshape(x, [0, -1])         x = self._fc1(x)         x = F.relu(x)         x = self._drop(x)         x = self._fc2(x)         x = F.relu(x)         x = self._drop(x)         x = self._out(x)         return x def VGG11(**args):     model = VGGNet(layers=11, **args)     return model def VGG13(**args):     model = VGGNet(layers=13, **args)     return model def VGG16(**args):     model = VGGNet(layers=16, **args)     return model def VGG19(**args):     model = VGGNet(layers=19, **args)     return model  
展开

社区升级了，直接粘代码不吃掉缩进了～～

0
回复
AIStudio810259
#9 回复于2021-02
AIStudio810258 #4
vgg就是直接码ConvBN

就想按照它那个写。。。

0
回复
AIStudio810259
#10 回复于2021-02
AIStudio810258 #7
import paddle from paddle import ParamAttr import paddle.nn as nn import paddle.nn.functional as F from paddle.nn import Conv2D, BatchNorm, Linear, Dropout from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D __all__ = ["VGG11", "VGG13", "VGG16", "VGG19"] class ConvBlock(nn.Layer):     def __init__(self, input_channels, output_channels, groups, name=None):         super(ConvBlock, self).__init__()         self.groups = groups         self._conv_1 = Conv2D(             in_channels=input_channels,             out_channels=output_channels,             kernel_size=3,             stride=1,             padding=1,             weight_attr=ParamAttr(name=name + "1_weights"),             bias_attr=False)         if groups == 2 or groups == 3 or groups == 4:             self._conv_2 = Conv2D(                 in_channels=output_channels,                 out_channels=output_channels,                 kernel_size=3,                 stride=1,                 padding=1,                 weight_attr=ParamAttr(name=name + "2_weights"),                 bias_attr=False)         if groups == 3 or groups == 4:             self._conv_3 = Conv2D(                 in_channels=output_channels,                 out_channels=output_channels,                 kernel_size=3,                 stride=1,                 padding=1,                 weight_attr=ParamAttr(name=name + "3_weights"),                 bias_attr=False)         if groups == 4:             self._conv_4 = Conv2D(                 in_channels=output_channels,                 out_channels=output_channels,                 kernel_size=3,                 stride=1,                 padding=1,                 weight_attr=ParamAttr(name=name + "4_weights"),                 bias_attr=False)         self._pool = MaxPool2D(kernel_size=2, stride=2, padding=0)     def forward(self, inputs):         x = self._conv_1(inputs)         x = F.relu(x)         if self.groups == 2 or self.groups == 3 or self.groups == 4:             x = self._conv_2(x)             x = F.relu(x)         if self.groups == 3 or self.groups == 4:             x = self._conv_3(x)             x = F.relu(x)         if self.groups == 4:             x = self._conv_4(x)             x = F.relu(x)         x = self._pool(x)         return x class VGGNet(nn.Layer):     def __init__(self, layers=11, class_dim=1000):         super(VGGNet, self).__init__()         self.layers = layers         self.vgg_configure = {             11: [1, 1, 2, 2, 2],             13: [2, 2, 2, 2, 2],             16: [2, 2, 3, 3, 3],             19: [2, 2, 4, 4, 4]         }         assert self.layers in self.vgg_configure.keys(), \             "supported layers are {} but input layer is {}".format(                 vgg_configure.keys(), layers)         self.groups = self.vgg_configure[self.layers]         self._conv_block_1 = ConvBlock(3, 64, self.groups[0], name="conv1_")         self._conv_block_2 = ConvBlock(64, 128, self.groups[1], name="conv2_")         self._conv_block_3 = ConvBlock(128, 256, self.groups[2], name="conv3_")         self._conv_block_4 = ConvBlock(256, 512, self.groups[3], name="conv4_")         self._conv_block_5 = ConvBlock(512, 512, self.groups[4], name="conv5_")         self._drop = Dropout(p=0.5, mode="downscale_in_infer")         self._fc1 = Linear(             7 * 7 * 512,             4096,             weight_attr=ParamAttr(name="fc6_weights"),             bias_attr=ParamAttr(name="fc6_offset"))         self._fc2 = Linear(             4096,             4096,             weight_attr=ParamAttr(name="fc7_weights"),             bias_attr=ParamAttr(name="fc7_offset"))         self._out = Linear(             4096,             class_dim,             weight_attr=ParamAttr(name="fc8_weights"),             bias_attr=ParamAttr(name="fc8_offset"))     def forward(self, inputs):         x = self._conv_block_1(inputs)         x = self._conv_block_2(x)         x = self._conv_block_3(x)         x = self._conv_block_4(x)         x = self._conv_block_5(x)         x = paddle.reshape(x, [0, -1])         x = self._fc1(x)         x = F.relu(x)         x = self._drop(x)         x = self._fc2(x)         x = F.relu(x)         x = self._drop(x)         x = self._out(x)         return x def VGG11(**args):     model = VGGNet(layers=11, **args)     return model def VGG13(**args):     model = VGGNet(layers=13, **args)     return model def VGG16(**args):     model = VGGNet(layers=16, **args)     return model def VGG19(**args):     model = VGGNet(layers=19, **args)     return model  
展开

---------------------------------------------------------------------------
Layer (type) Input Shape Output Shape Param #
===========================================================================
Conv2D-1 [[1, 3, 224, 224]] [1, 64, 224, 224] 1,792
Conv2D-2 [[1, 64, 224, 224]] [1, 64, 224, 224] 36,928
MaxPool2D-1 [[1, 64, 224, 224]] [1, 64, 112, 112] 0
ConvPool-1 [[1, 3, 224, 224]] [1, 64, 112, 112] 0
Conv2D-3 [[1, 64, 112, 112]] [1, 128, 112, 112] 73,856
Conv2D-4 [[1, 128, 112, 112]] [1, 128, 112, 112] 147,584
MaxPool2D-2 [[1, 128, 112, 112]] [1, 128, 56, 56] 0
ConvPool-2 [[1, 64, 112, 112]] [1, 128, 56, 56] 0
Conv2D-5 [[1, 128, 56, 56]] [1, 256, 56, 56] 295,168
Conv2D-6 [[1, 256, 56, 56]] [1, 256, 56, 56] 590,080
Conv2D-7 [[1, 256, 56, 56]] [1, 256, 56, 56] 590,080
MaxPool2D-3 [[1, 256, 56, 56]] [1, 256, 28, 28] 0
ConvPool-3 [[1, 128, 56, 56]] [1, 256, 28, 28] 0
Conv2D-8 [[1, 256, 28, 28]] [1, 512, 28, 28] 1,180,160
Conv2D-9 [[1, 512, 28, 28]] [1, 512, 28, 28] 2,359,808
Conv2D-10 [[1, 512, 28, 28]] [1, 512, 28, 28] 2,359,808
MaxPool2D-4 [[1, 512, 28, 28]] [1, 512, 14, 14] 0
ConvPool-4 [[1, 256, 28, 28]] [1, 512, 14, 14] 0
Conv2D-11 [[1, 512, 14, 14]] [1, 512, 14, 14] 2,359,808
Conv2D-12 [[1, 512, 14, 14]] [1, 512, 14, 14] 2,359,808
Conv2D-13 [[1, 512, 14, 14]] [1, 512, 14, 14] 2,359,808
MaxPool2D-5 [[1, 512, 14, 14]] [1, 512, 7, 7] 0
ConvPool-5 [[1, 512, 14, 14]] [1, 512, 7, 7] 0
Linear-1 [[1, 25088]] [1, 4096] 102,764,544
Linear-2 [[1, 4096]] [1, 4096] 16,781,312
Linear-3 [[1, 4096]] [1, 5] 20,485
===========================================================================
Total params: 134,281,029
Trainable params: 134,281,029
Non-trainable params: 0
---------------------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 126.77
Params size (MB): 512.24
Estimated Total Size (MB): 639.59
---------------------------------------------------------------------------


{'total_params': 134281029, 'trainable_params': 134281029}

 

 

 

0
回复
AIStudio810259
#11 回复于2021-02

Epoch 9/10
step 48/48 [==============================] - loss: 1.1617 - acc_top1: 0.7626 - acc_top2: 0.8937 - 849ms/step
save checkpoint at /home/aistudio/chk_points/8
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 7/7 [==============================] - loss: 1.1522 - acc_top1: 0.7727 - acc_top2: 0.8955 - 624ms/step
Eval samples: 440
Epoch 10/10
step 48/48 [==============================] - loss: 1.1092 - acc_top1: 0.7750 - acc_top2: 0.9003 - 846ms/step
save checkpoint at /home/aistudio/chk_points/9
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 7/7 [==============================] - loss: 1.1537 - acc_top1: 0.7864 - acc_top2: 0.9045 - 617ms/step
Eval samples: 440
save checkpoint at /home/aistudio/chk_points/final

 
 
 

0
回复
AIStudio810259
#12 回复于2021-02

1
样本ID：2, 真实标签：river, 预测值：river
2
样本ID：38, 真实标签：lawn, 预测值：river
0
样本ID：56, 真实标签：ice, 预测值：ice
2
样本ID：92, 真实标签：lawn, 预测值：lawn
0
样本ID：100, 真实标签：ice, 预测值：river
1
样本ID：303, 真实标签：river, 预测值：river

0
回复
AIStudio810258
#13 回复于2021-02
AIStudio810259 #11
Epoch 9/10 step 48/48 [==============================] - loss: 1.1617 - acc_top1: 0.7626 - acc_top2: 0.8937 - 849ms/step save checkpoint at /home/aistudio/chk_points/8 Eval begin... The loss value printed in the log is the current batch, and the metric is the average value of previous step. step 7/7 [==============================] - loss: 1.1522 - acc_top1: 0.7727 - acc_top2: 0.8955 - 624ms/step Eval samples: 440 Epoch 10/10 step 48/48 [==============================] - loss: 1.1092 - acc_top1: 0.7750 - acc_top2: 0.9003 - 846ms/step save checkpoint at /home/aistudio/chk_points/9 Eval begin... The loss value printed in the log is the current batch, and the metric is the average value of previous step. step 7/7 [==============================] - loss: 1.1537 - acc_top1: 0.7864 - acc_top2: 0.9045 - 617ms/step Eval samples: 440 save checkpoint at /home/aistudio/chk_points/final      
展开

这是直接用paddleclas训的吧，效果不错啊

0
回复
AIStudio810258
#14 回复于2021-02
AIStudio810259 #10
--------------------------------------------------------------------------- Layer (type) Input Shape Output Shape Param # =========================================================================== Conv2D-1 [[1, 3, 224, 224]] [1, 64, 224, 224] 1,792 Conv2D-2 [[1, 64, 224, 224]] [1, 64, 224, 224] 36,928 MaxPool2D-1 [[1, 64, 224, 224]] [1, 64, 112, 112] 0 ConvPool-1 [[1, 3, 224, 224]] [1, 64, 112, 112] 0 Conv2D-3 [[1, 64, 112, 112]] [1, 128, 112, 112] 73,856 Conv2D-4 [[1, 128, 112, 112]] [1, 128, 112, 112] 147,584 MaxPool2D-2 [[1, 128, 112, 112]] [1, 128, 56, 56] 0 ConvPool-2 [[1, 64, 112, 112]] [1, 128, 56, 56] 0 Conv2D-5 [[1, 128, 56, 56]] [1, 256, 56, 56] 295,168 Conv2D-6 [[1, 256, 56, 56]] [1, 256, 56, 56] 590,080 Conv2D-7 [[1, 256, 56, 56]] [1, 256, 56, 56] 590,080 MaxPool2D-3 [[1, 256, 56, 56]] [1, 256, 28, 28] 0 ConvPool-3 [[1, 128, 56, 56]] [1, 256, 28, 28] 0 Conv2D-8 [[1, 256, 28, 28]] [1, 512, 28, 28] 1,180,160 Conv2D-9 [[1, 512, 28, 28]] [1, 512, 28, 28] 2,359,808 Conv2D-10 [[1, 512, 28, 28]] [1, 512, 28, 28] 2,359,808 MaxPool2D-4 [[1, 512, 28, 28]] [1, 512, 14, 14] 0 ConvPool-4 [[1, 256, 28, 28]] [1, 512, 14, 14] 0 Conv2D-11 [[1, 512, 14, 14]] [1, 512, 14, 14] 2,359,808 Conv2D-12 [[1, 512, 14, 14]] [1, 512, 14, 14] 2,359,808 Conv2D-13 [[1, 512, 14, 14]] [1, 512, 14, 14] 2,359,808 MaxPool2D-5 [[1, 512, 14, 14]] [1, 512, 7, 7] 0 ConvPool-5 [[1, 512, 14, 14]] [1, 512, 7, 7] 0 Linear-1 [[1, 25088]] [1, 4096] 102,764,544 Linear-2 [[1, 4096]] [1, 4096] 16,781,312 Linear-3 [[1, 4096]] [1, 5] 20,485 =========================================================================== Total params: 134,281,029 Trainable params: 134,281,029 Non-trainable params: 0 --------------------------------------------------------------------------- Input size (MB): 0.57 Forward/backward pass size (MB): 126.77 Params size (MB): 512.24 Estimated Total Size (MB): 639.59 --------------------------------------------------------------------------- {'total_params': 134281029, 'trainable_params': 134281029}      
展开

vgg19

0
回复
AIStudio810259
#15 回复于2021-02
AIStudio810258 #13
这是直接用paddleclas训的吧，效果不错啊

手写的

0
回复
AIStudio810258
#16 回复于2021-02
AIStudio810259 #15
手写的

打印风格太paddleclas了～

0
回复
AIStudio810259
#17 回复于2021-02
AIStudio810258 #16
打印风格太paddleclas了～

反正看图识字描完了。

0
回复
七年期限
#18 回复于2021-02

没有现成得吗？

0
回复
AIStudio810259
#19 回复于2021-02
七年期限 #18
没有现成得吗？

现成的没有现做的香

0
回复
七年期限
#20 回复于2021-02
AIStudio810259 #19
现成的没有现做的香

哈哈 那倒是

0
回复
七年期限
#21 回复于2021-02

根据他那个 改改

0
回复
关于AI Studio
AI Studio是基于百度深度学习平台飞桨的人工智能学习与实训社区，提供在线编程环境、免费GPU算力、海量开源算法和开放数据，帮助开发者快速创建和部署模型。
了解
相关资源
用户指南 常见问题
教育专区
教育版介绍 教育版使用文档 教师开课申请
联系我们
邮箱: aistudio@baidu.com 官方QQ：580959619
飞桨微信服务号
友情链接：
飞桨官网
|
飞桨源码
|
百度开发者中心
|
百度云智学院
|
百度技术学院
|
百度效率云
|
百度点石
|
用户协议
|
© 使用百度前必读
提issue
需求/bug反馈？一键提issue告诉我们
提pr
发现bug？如果您知道修复办法，欢迎提pr直接参与建设飞桨~
在@后输入用户全名并按空格结束，可艾特全站任一用户