X2Paddle实践之——Tensorflow版本VGG模型转化为paddle版本模型

╰半夏微凉° 2022-10-17 13:51 119阅读 0赞

最近经常会刷到很多关于paddle的文章，说实在的paddle我之前读研的时候还是有用过的，后面就没怎么再用过了，经过几年的捶打完善，感觉现在已经是一款不错的深度学习框架了，是时候捡起来学习了解一下了。

PaddlePaddle地址在[这里][Link 1]，首页截图如下所示：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70][]

上面图中的每一个项目都是当前很火的项目，不过我们今天的主要学习内容不在上面，之前也做过不同框架之间模型转化的工作，今天也是这样的内容，因为PaddlePaddle自家提供了类似的开源工具X2Paddle，顾名思义，就是提供其他框架转化为PaddlePaddle自家模型文件的工具。

话不多说，这里先来看X2Paddle，地址在[这里][Link 2]，首页如下：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 1][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 2][]

感兴趣的话可以去仔细了解一下，我们今天主要是借助具体的实践来学习一下具体的操作使用，VGG\_16是CV领域的一个经典模型，这里就以VGG\_16为例，给大家展示如何将TensorFlow训练好的模型转换为飞桨模型。

首先下载模型，地址如下：

http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz

之后解压缩，截图如下所示：

![20210603162945408.png][]

接下来我们需要重新加载参数，并将网络结构和参数一起保存为checkpoint模型，代码如下：

with tf.Session() as sess:
        inputs = tf.placeholder(dtype=tf.float32, shape=[None, 224, 224, 3], name="inputs")
        logits, endpoint = vgg.vgg_16(inputs, num_classes=1000, is_training=False)
        load_model = slim.assign_from_checkpoint_fn("vgg_16.ckpt", slim.get_model_variables("vgg_16"))
        load_model(sess)
        data = numpy.random.rand(5, 224, 224, 3)
        input_tensor = sess.graph.get_tensor_by_name("inputs:0")
        output_tensor = sess.graph.get_tensor_by_name("vgg_16/fc8/squeezed:0")
        result = sess.run([output_tensor], {input_tensor:data})
        numpy.save("tensorflow.npy", numpy.array(result))
        saver = tf.train.Saver()
        saver.save(sess, "checkpoint/model")

完成上述处理，我们就可以着手将模型转为PaddlePaddle模型了，代码如下：

parser = convert._get_parser()
    parser.meta_file = "checkpoint/model.meta"
    parser.ckpt_dir = "checkpoint"
    parser.in_nodes = ["inputs"]
    parser.input_shape = ["None,224,224,3"]
    parser.output_nodes = ["vgg_16/fc8/squeezed"]
    parser.use_cuda = "True"
    parser.input_format = "NHWC"
    parser.save_dir = "paddle_model"
    convert.run(parser)

由于是初次学习使用，这里我附上完整的日志输出，如下：

Use tf.compat.v1.graph_util.extract_sub_graph
    INFO:root:Tensorflow model loaded!
    INFO:root:TotalNum:86,TraslatedNum:1,CurrentNode:inputs
    INFO:root:TotalNum:86,TraslatedNum:2,CurrentNode:vgg_16/conv1/conv1_1/weights
    INFO:root:TotalNum:86,TraslatedNum:3,CurrentNode:vgg_16/conv1/conv1_1/biases
    INFO:root:TotalNum:86,TraslatedNum:4,CurrentNode:vgg_16/conv1/conv1_2/weights
    INFO:root:TotalNum:86,TraslatedNum:5,CurrentNode:vgg_16/conv1/conv1_2/biases
    INFO:root:TotalNum:86,TraslatedNum:6,CurrentNode:vgg_16/conv2/conv2_1/weights
    INFO:root:TotalNum:86,TraslatedNum:7,CurrentNode:vgg_16/conv2/conv2_1/biases
    INFO:root:TotalNum:86,TraslatedNum:8,CurrentNode:vgg_16/conv2/conv2_2/weights
    INFO:root:TotalNum:86,TraslatedNum:9,CurrentNode:vgg_16/conv2/conv2_2/biases
    INFO:root:TotalNum:86,TraslatedNum:10,CurrentNode:vgg_16/conv3/conv3_1/weights
    INFO:root:TotalNum:86,TraslatedNum:11,CurrentNode:vgg_16/conv3/conv3_1/biases
    INFO:root:TotalNum:86,TraslatedNum:12,CurrentNode:vgg_16/conv3/conv3_2/weights
    INFO:root:TotalNum:86,TraslatedNum:13,CurrentNode:vgg_16/conv3/conv3_2/biases
    INFO:root:TotalNum:86,TraslatedNum:14,CurrentNode:vgg_16/conv3/conv3_3/weights
    INFO:root:TotalNum:86,TraslatedNum:15,CurrentNode:vgg_16/conv3/conv3_3/biases
    INFO:root:TotalNum:86,TraslatedNum:16,CurrentNode:vgg_16/conv4/conv4_1/weights
    INFO:root:TotalNum:86,TraslatedNum:17,CurrentNode:vgg_16/conv4/conv4_1/biases
    INFO:root:TotalNum:86,TraslatedNum:18,CurrentNode:vgg_16/conv4/conv4_2/weights
    INFO:root:TotalNum:86,TraslatedNum:19,CurrentNode:vgg_16/conv4/conv4_2/biases
    INFO:root:TotalNum:86,TraslatedNum:20,CurrentNode:vgg_16/conv4/conv4_3/weights
    INFO:root:TotalNum:86,TraslatedNum:21,CurrentNode:vgg_16/conv4/conv4_3/biases
    INFO:root:TotalNum:86,TraslatedNum:22,CurrentNode:vgg_16/conv5/conv5_1/weights
    INFO:root:TotalNum:86,TraslatedNum:23,CurrentNode:vgg_16/conv5/conv5_1/biases
    INFO:root:TotalNum:86,TraslatedNum:24,CurrentNode:vgg_16/conv5/conv5_2/weights
    INFO:root:TotalNum:86,TraslatedNum:25,CurrentNode:vgg_16/conv5/conv5_2/biases
    INFO:root:TotalNum:86,TraslatedNum:26,CurrentNode:vgg_16/conv5/conv5_3/weights
    INFO:root:TotalNum:86,TraslatedNum:27,CurrentNode:vgg_16/conv5/conv5_3/biases
    INFO:root:TotalNum:86,TraslatedNum:28,CurrentNode:vgg_16/fc6/weights
    INFO:root:TotalNum:86,TraslatedNum:29,CurrentNode:vgg_16/fc6/biases
    INFO:root:TotalNum:86,TraslatedNum:30,CurrentNode:vgg_16/fc7/weights
    INFO:root:TotalNum:86,TraslatedNum:31,CurrentNode:vgg_16/fc7/biases
    INFO:root:TotalNum:86,TraslatedNum:32,CurrentNode:vgg_16/fc8/weights
    INFO:root:TotalNum:86,TraslatedNum:33,CurrentNode:vgg_16/fc8/biases
    INFO:root:TotalNum:86,TraslatedNum:34,CurrentNode:vgg_16/conv1/conv1_1/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:35,CurrentNode:vgg_16/conv1/conv1_1/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:36,CurrentNode:vgg_16/conv1/conv1_1/Relu
    INFO:root:TotalNum:86,TraslatedNum:37,CurrentNode:vgg_16/conv1/conv1_2/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:38,CurrentNode:vgg_16/conv1/conv1_2/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:39,CurrentNode:vgg_16/conv1/conv1_2/Relu
    INFO:root:TotalNum:86,TraslatedNum:40,CurrentNode:vgg_16/pool1/MaxPool
    INFO:root:TotalNum:86,TraslatedNum:41,CurrentNode:vgg_16/conv2/conv2_1/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:42,CurrentNode:vgg_16/conv2/conv2_1/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:43,CurrentNode:vgg_16/conv2/conv2_1/Relu
    INFO:root:TotalNum:86,TraslatedNum:44,CurrentNode:vgg_16/conv2/conv2_2/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:45,CurrentNode:vgg_16/conv2/conv2_2/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:46,CurrentNode:vgg_16/conv2/conv2_2/Relu
    INFO:root:TotalNum:86,TraslatedNum:47,CurrentNode:vgg_16/pool2/MaxPool
    INFO:root:TotalNum:86,TraslatedNum:48,CurrentNode:vgg_16/conv3/conv3_1/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:49,CurrentNode:vgg_16/conv3/conv3_1/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:50,CurrentNode:vgg_16/conv3/conv3_1/Relu
    INFO:root:TotalNum:86,TraslatedNum:51,CurrentNode:vgg_16/conv3/conv3_2/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:52,CurrentNode:vgg_16/conv3/conv3_2/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:53,CurrentNode:vgg_16/conv3/conv3_2/Relu
    INFO:root:TotalNum:86,TraslatedNum:54,CurrentNode:vgg_16/conv3/conv3_3/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:55,CurrentNode:vgg_16/conv3/conv3_3/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:56,CurrentNode:vgg_16/conv3/conv3_3/Relu
    INFO:root:TotalNum:86,TraslatedNum:57,CurrentNode:vgg_16/pool3/MaxPool
    INFO:root:TotalNum:86,TraslatedNum:58,CurrentNode:vgg_16/conv4/conv4_1/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:59,CurrentNode:vgg_16/conv4/conv4_1/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:60,CurrentNode:vgg_16/conv4/conv4_1/Relu
    INFO:root:TotalNum:86,TraslatedNum:61,CurrentNode:vgg_16/conv4/conv4_2/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:62,CurrentNode:vgg_16/conv4/conv4_2/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:63,CurrentNode:vgg_16/conv4/conv4_2/Relu
    INFO:root:TotalNum:86,TraslatedNum:64,CurrentNode:vgg_16/conv4/conv4_3/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:65,CurrentNode:vgg_16/conv4/conv4_3/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:66,CurrentNode:vgg_16/conv4/conv4_3/Relu
    INFO:root:TotalNum:86,TraslatedNum:67,CurrentNode:vgg_16/pool4/MaxPool
    INFO:root:TotalNum:86,TraslatedNum:68,CurrentNode:vgg_16/conv5/conv5_1/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:69,CurrentNode:vgg_16/conv5/conv5_1/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:70,CurrentNode:vgg_16/conv5/conv5_1/Relu
    INFO:root:TotalNum:86,TraslatedNum:71,CurrentNode:vgg_16/conv5/conv5_2/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:72,CurrentNode:vgg_16/conv5/conv5_2/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:73,CurrentNode:vgg_16/conv5/conv5_2/Relu
    INFO:root:TotalNum:86,TraslatedNum:74,CurrentNode:vgg_16/conv5/conv5_3/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:75,CurrentNode:vgg_16/conv5/conv5_3/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:76,CurrentNode:vgg_16/conv5/conv5_3/Relu
    INFO:root:TotalNum:86,TraslatedNum:77,CurrentNode:vgg_16/pool5/MaxPool
    INFO:root:TotalNum:86,TraslatedNum:78,CurrentNode:vgg_16/fc6/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:79,CurrentNode:vgg_16/fc6/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:80,CurrentNode:vgg_16/fc6/Relu
    INFO:root:TotalNum:86,TraslatedNum:81,CurrentNode:vgg_16/fc7/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:82,CurrentNode:vgg_16/fc7/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:83,CurrentNode:vgg_16/fc7/Relu
    INFO:root:TotalNum:86,TraslatedNum:84,CurrentNode:vgg_16/fc8/Conv2D
    INFO:root:TotalNum:86,TraslatedNum:85,CurrentNode:vgg_16/fc8/BiasAdd
    INFO:root:TotalNum:86,TraslatedNum:86,CurrentNode:vgg_16/fc8/squeezed
    INFO:root:Model translated!

执行成功截图如下所示：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 3][]

转化成功后会出现 paddle\_model 目录，如下：

![20210603163549713.png][]

截图如下所示：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 4][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 5][]

转化的工作到这里其实就已经结束了，但是一般的转化都逃不开精度损失的问题，这里需要做一下验证计算。代码如下：

model = ml.ModelLoader("paddle_model", use_cuda=False)
    numpy.random.seed(13)
    data = numpy.random.rand(5, 224, 224, 3).astype("float32")
    # NHWC -> NCHW
    data = numpy.transpose(data, (0, 3, 1, 2))
    results = model.inference(feed_dict={model.inputs[0]:data})
    numpy.save("paddle.npy", numpy.array(results))

执行的时候报错了，错误信息如下：

AssertionError: In PaddlePaddle 2.x, we turn on dynamic graph mode by default, and 'data()' is only supported in static graph mode. So if you want to use this api, please call 'paddle.enable_static()' before this api to enter static graph mode.

代码更正为下面的，其实主要就是静态图的问题。

paddle.enable_static()
    model = ml.ModelLoader("paddle_model", use_cuda=False)
    numpy.random.seed(13)
    data = numpy.random.rand(5, 224, 224, 3).astype("float32")
    # NHWC -> NCHW
    data = numpy.transpose(data, (0, 3, 1, 2))
    results = model.inference(feed_dict={model.inputs[0]:data})
    numpy.save("paddle.npy", numpy.array(results))

之后计算一下模型的损失，通过把两个模型文件加载进来后，通过numpy.fabs来求两个模型结果的差异即可，如下：

paddle_result = numpy.load("paddle.npy")
    tensorflow_result = numpy.load("tensorflow.npy")
    diff = numpy.fabs(paddle_result - tensorflow_result)
    print(numpy.max(diff))

结果如下所示：

![20210603163950947.png][]

误差的数据还是比较小的，在可接受的范围内。

最后附上完整代码，如下：

#!usr/bin/env python
    #encoding:utf-8
    from __future__ import division
    
    '''
    __Author__:沂水寒城
    功能： VGG Tensorflow 转 Paddle 实战
    '''
    
    
    import os
    import tensorflow.contrib.slim as slim
    from tensorflow.contrib.slim.nets import vgg
    import tensorflow as tf
    import numpy
    
    
    
    
    '''
    保存模型为checkpoint格式
    '''
    with tf.Session() as sess:
        inputs = tf.placeholder(dtype=tf.float32, shape=[None, 224, 224, 3], name="inputs")
        logits, endpoint = vgg.vgg_16(inputs, num_classes=1000, is_training=False)
        load_model = slim.assign_from_checkpoint_fn("vgg_16.ckpt", slim.get_model_variables("vgg_16"))
        load_model(sess)
        data = numpy.random.rand(5, 224, 224, 3)
        input_tensor = sess.graph.get_tensor_by_name("inputs:0")
        output_tensor = sess.graph.get_tensor_by_name("vgg_16/fc8/squeezed:0")
        result = sess.run([output_tensor], {input_tensor:data})
        numpy.save("tensorflow.npy", numpy.array(result))
        saver = tf.train.Saver()
        saver.save(sess, "checkpoint/model")
    
    
    
    '''
    将模型转换为飞桨模型
    '''
    import tf2fluid.convert as convert
    import argparse
    parser = convert._get_parser()
    parser.meta_file = "checkpoint/model.meta"
    parser.ckpt_dir = "checkpoint"
    parser.in_nodes = ["inputs"]
    parser.input_shape = ["None,224,224,3"]
    parser.output_nodes = ["vgg_16/fc8/squeezed"]
    parser.use_cuda = "True"
    parser.input_format = "NHWC"
    parser.save_dir = "paddle_model"
    convert.run(parser)
    
    
    
    
    '''
    预测结果差异对比
    '''
    import numpy
    import paddle
    import tf2fluid.model_loader as ml
    paddle.enable_static()
    model = ml.ModelLoader("paddle_model", use_cuda=False)
    numpy.random.seed(13)
    data = numpy.random.rand(5, 224, 224, 3).astype("float32")
    # NHWC -> NCHW
    data = numpy.transpose(data, (0, 3, 1, 2))
    results = model.inference(feed_dict={model.inputs[0]:data})
    numpy.save("paddle.npy", numpy.array(results))
    
    
    
    '''
    对比模型损失
    '''
    import numpy
    paddle_result = numpy.load("paddle.npy")
    tensorflow_result = numpy.load("tensorflow.npy")
    diff = numpy.fabs(paddle_result - tensorflow_result)
    print(numpy.max(diff))

初次的实践使用就到这里了，后面有时间继续深入研究下。

[Link 1]: https://github.com/PaddlePaddle
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70]: /images/20221014/344db4f8503445c0ba4148a2ecb8e7ea.png
[Link 2]: https://github.com/PaddlePaddle/X2Paddle
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 1]: /images/20221014/02fac5ae322846b5a77e8b74853d2de0.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 2]: /images/20221014/9b137766ea094ae789e654679e59145a.png
[20210603162945408.png]: /images/20221014/6179972c2dab484990e782f5aedcf897.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 3]: /images/20221014/270634c88ec54b369bb0a0e2b0287d4d.png
[20210603163549713.png]: /images/20221014/20587637c9354237bb6e3a491cb5495d.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 4]: /images/20221014/f72f002a20c2436d976ed5c997a3d09e.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1RvZ2V0aGVyX0Na_size_16_color_FFFFFF_t_70 5]: /images/20221014/cad8a9f9947b4c069f88e708f2303d0b.png
[20210603163950947.png]: /images/20221014/ed5f8850206e42eb98d0bd84659f6729.png