微软与Nvidia、AMD 和英特尔等供应商合作，以确保在 Windows 10及以上和 WSL 上提供流畅的体验，以便在支持 DirectX 12 的 GPU 上加速训练。关于GPU的驱动支持，请看微软的说明：TensorFlow with DirectML on WSL | Microsoft Docs

2.Python 环境及Tensorflow-DirecTML安装注意事项

我使用conda 作为python环境管理，非常方便，简单易用,安装成功后，应当创建专用于DirecTML的环境，我是这样的：

conda activate directm

安装DirecTML：

pip install tensorflow-directml

注意：如果原先有安装过 tensorflow，一定要卸载,如果又有安装过tensorflow-directml，而且未起作用，也应当一并卸载。

pip uninstall tensorflow

pip uninstall tensorflow-directml

如果不是新创建的conda directml，尽量卸载完所有tensorflow有关的模块,否则。。。。

pip list | grep tensorflow

pip uninstall tensorflow***

安装成功后，查看下tensorflow相关包，应当是这个样子的，注意：不能再去安装 tensorflow：

pip list | grep tensor

tensorboard 1.15.0
tensorboard-data-server 0.6.1
tensorboard-plugin-wit 1.8.1
tensorflow-directml 1.15.5
tensorflow-estimator 1.15.1

3.目标识别输入数据处理

将已经标定的图片及xml文件转换成 .record文件（xml文件可以使用labelImg进行标识产生，不在本文讨论范围。）：

首先是 xml转换为csv，代码中涉及路径的，需要根据实际情况修改,python xml_to_csv.py中有一处需要修改：


cd /home/akun-u/dev/deeplearning/raccoon_dataset
 
python xml_to_csv.py

执行上述命令后，会在当前目录生成两个csv文件

Successfully converted files:201 xml to csv(./train_labels.csv).
Successfully converted files:16 xml to csv(./test_labels.csv).

xml_to_csv.py文件内容如下：


import glob
import os
import xml.etree.ElementTree as ET
 
import pandas as pd
 
 
def xml_to_csv(path):
    xml_list = []
    for xml_file in glob.glob(path + '/*.xml'):
        tree = ET.parse(xml_file)
        root = tree.getroot()
        for member in root.findall('object'):
            value = (root.find('filename').text,
                     int(root.find('size')[0].text),
                     int(root.find('size')[1].text),
                     member[0].text,
                     int(member[4][0].text),
                     int(member[4][1].text),
                     int(member[4][2].text),
                     int(member[4][3].text)
                     )
            xml_list.append(value)
    column_name = ['filename', 'width', 'height', 'class', 'xmin', 'ymin', 'xmax', 'ymax']
    xml_df = pd.DataFrame(xml_list, columns=column_name)
    return xml_df
 
def main():
    for direcory in ['train','test']:
        image_path = os.path.join('/home/akun-u/dev/deeplearning/raccoon_dataset/annotations', direcory)
        print("xmlfile:", image_path)
        xml_df = xml_to_csv(image_path)
 
        csv_path = './{0}_labels.csv'.format(direcory)
        xml_df.to_csv(csv_path, index=None)
        print(f'Successfully converted files:{len(xml_df)} xml to csv({csv_path}).')
 
main()
'运行

接着将csv文件转换为 .record:注意文件路径要填写正确。


cd /home/akun-u/dev/deeplearning/raccoon_dataset
 
python generate_tfrecord.py --csv_input=test_labels.csv --output_path=data/dataset/test.record --image_dir=images
 
python generate_tfrecord.py --csv_input=train_labels.csv --output_path=data/dataset/train.record --image_dir=images

generate_tfrecord.py文件内容如下：


"""
Usage:
  # From tensorflow/models/
  # Create train data:
  python generate_tfrecord.py --csv_input=data/train_labels.csv  --output_path=train.record
  # Create test data:
  python generate_tfrecord.py --csv_input=data/test_labels.csv  --output_path=test.record
#yhk add
python generate_tfrecord.py \
    --csv_input=data/test_labels.csv \
    --output_path=data/dataset/test.record \
    --image_dir=images/
python generate_tfrecord.py \
    --csv_input=data/train_labels.csv \
    --output_path=data/dataset/train.record \
    --image_dir=images/
"""
from __future__ import absolute_import, division, print_function
 
import io
import os
from collections import OrderedDict, namedtuple
 
import pandas as pd
import tensorflow as tf2
 
tf= tf2.compat.v1 #yhk add
 
from object_detection.utils import dataset_util
from PIL import Image
 
flags = tf.app.flags
flags.DEFINE_string('csv_input', '', 'Path to the CSV input')
flags.DEFINE_string('output_path', '', 'Path to output TFRecord')
flags.DEFINE_string('image_dir', '', 'Path to images')
FLAGS = flags.FLAGS
 
 
# TO-DO replace this with label map
def class_text_to_int(row_label):
    if row_label == 'raccoon':
        return 1
    else:
        return 2
 
 
def split(df, group):
    data = namedtuple('data', ['filename', 'object'])
    gb = df.groupby(group)
    return [data(filename, gb.get_group(x)) for filename, x in zip(gb.groups.keys(), gb.groups)]
 
 
def create_tf_example(group, path):
    with tf.gfile.GFile(os.path.join(path, '{}'.format(group.filename)), 'rb') as fid:
        encoded_jpg = fid.read()
    encoded_jpg_io = io.BytesIO(encoded_jpg)
    image = Image.open(encoded_jpg_io)
    width, height = image.size
 
    filename = group.filename.encode('utf8')
    image_format = b'jpg'
    xmins = []
    xmaxs = []
    ymins = []
    ymaxs = []
    classes_text = []
    classes = []
 
    for index, row in group.object.iterrows():
        xmins.append(row['xmin'] / width)
        xmaxs.append(row['xmax'] / width)
        ymins.append(row['ymin'] / height)
        ymaxs.append(row['ymax'] / height)
        classes_text.append(row['class'].encode('utf8'))
        classes.append(class_text_to_int(row['class']))
 
    tf_example = tf.train.Example(features=tf.train.Features(feature={
        'image/height': dataset_util.int64_feature(height),
        'image/width': dataset_util.int64_feature(width),
        'image/filename': dataset_util.bytes_feature(filename),
        'image/source_id': dataset_util.bytes_feature(filename),
        'image/encoded': dataset_util.bytes_feature(encoded_jpg),
        'image/format': dataset_util.bytes_feature(image_format),
        'image/object/bbox/xmin': dataset_util.float_list_feature(xmins),
        'image/object/bbox/xmax': dataset_util.float_list_feature(xmaxs),
        'image/object/bbox/ymin': dataset_util.float_list_feature(ymins),
        'image/object/bbox/ymax': dataset_util.float_list_feature(ymaxs),
        'image/object/class/text': dataset_util.bytes_list_feature(classes_text),
        'image/object/class/label': dataset_util.int64_list_feature(classes),
    }))
    return tf_example
 
 
def main(_):
    writer = tf.python_io.TFRecordWriter(FLAGS.output_path)
    path = os.path.join(FLAGS.image_dir)
    examples = pd.read_csv(FLAGS.csv_input)
    grouped = split(examples, 'filename')
    for group in grouped:
        tf_example = create_tf_example(group, path)
        writer.write(tf_example.SerializeToString())
 
    writer.close()
    output_path = os.path.join(os.getcwd(), FLAGS.output_path)
    print('Successfully created the TFRecords: {}'.format(output_path))
 
 
if __name__ == '__main__':
    tf.app.run()

4.进行数据处理，学习

在接着执行以下语句之前，请：

1.修改 /deeplearning/pipeline.config 文件中的路径（有5处）

2.根据实际情况调整 /deeplearning/conf/clickv2_label_map.pbtxt

接下来执行：python object_detection/model_main.py

注意：执行python object_detection/model_main.py时，尽量一行写完整个语句，我原先使用了 \ 连接多行语句，一直调试失败，浪费了不少时间。

语句执行过程中，有出现 “DirectML: creating device on adapter 0 (AMD Radeon RX 6700 XT”，字样

RX 6700 XT是我的显卡型号，请注意看，该字样能够确定是否有使用GPU，这个执行过程会花费很长时间，我的显卡在--num_train_steps=50000 --num_eval_steps=2000的情况下，执行了约3小时。


cd  /home/akun-u/dev/deeplearning/models/research
 
python object_detection/model_main.py --pipeline_config_path=/home/akun-u/dev/deeplearning/conf/ssd_mobilenet_v2_coco_2018_03_29/pipeline.config --model_dir=/home/akun-u/dev/deeplearning/models --num_train_steps=50000 --num_eval_steps=2000 --alsologtostderr

该命令会在 models目录下生成多个model.ckpt开头的文件。

执行情况如图：

5.生成PB文件，用于后续的目标识别


cd /home/akun-u/dev/deeplearning/models/research/object_detection
 
python export_inference_graph.py --input_type=image_tensor --pipeline_config_path=/home/akun-u/dev/deeplearning/conf/ssd_mobilenet_v2_coco_2018_03_29/pipeline.config --trained_checkpoint_prefix=/home/akun-u/dev/deeplearning/models/model.ckpt-50000 --output_directory=/home/akun-u/dev/deeplearning/train_pb

6.目标识别

注意：修改识别后文件保存路径（img_recongnize.py中：1处）


cd /home/akun-u/dev/deeplearning
 
python img_recongnize.py

运行后会在指定目录下生成多个jpg文件，打开可看到可识别目标被框选，如下图：

总结

以上就是今天要讲的内容，本文仅仅简单介绍了tensorflow-directmp的基本使用，自己过了好多坑，终于爬出，留个文本。。。

我的代码：https://download.csdn.net/download/sinat_32065311/85275745

代码来源：

GitHub - microsoft/tensorflow-directml: Fork of TensorFlow accelerated by DirectML

https://github.com/tensorflow/models

https://github.com/datitran/raccoon_dataset

Tensorflow Object detection 教程

声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/寸_铁/article/detail/938837