基於ssd的手勢識別模型（object detection api方式）

【Tensorflow】Object Detection API-訓練自己的手勢識別模型

1. 安裝tensorflow以及下載object detection api

1.安裝tensorflow：

對於CPU版本：pip install tensorflow
對於GPU版本：pip install tensorflow-gpu
升級tensorflow到最新版1.4.0：pip install --upgrade tensorflow-gpu

2.安裝必須庫：

sudo pip install pillow
sudo pip install lxml
sudo pip install jupyter
sudo pip install matplotlib
3.下載object detection api:
t clone https://github.com/tensorflow/models.git
4.protobuf編譯：在tensorflow/models/research/目錄下
protoc object_detection/protos/*.proto --python_out=.
5.添加pythonpath,在tensorflow/models/research/目錄下
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim
6.測試安裝：
python object_detection/builders/model_builder_test.py

2.訓練數據集准備

1.在model下新建文件夾dataset/VOCdevkit/VOC2007目錄，VOC2007目錄機構如下圖所示：

2.在VOC2007目錄下運行generate_txt.py程序，划分數據集，我的訓練集和驗證集比例為7:3，總量為1557

import os  
import random  
  
train_percent = 0.70
xmlfilepath = 'Annotations'  
txtsavepath = 'ImageSets\Main'  
total_xml = os.listdir(xmlfilepath)  
  
num=len(total_xml)  
list=range(num)  
tr=int(num*train_percent)  
train=random.sample(list,tr)  
   
ftrain = open('ImageSets/Main/train.txt', 'w')  
fval = open('ImageSets/Main/val.txt', 'w')  
  
for i  in list:  
    name=total_xml[i][:-4]+'\n'   
    if i in train:  
        ftrain.write(name)  
    else:  
        fval.write(name)  
   
ftrain.close()  
fval.close()   
print ("finished")

3.將models/research/object_detection/dataset_tools目錄下的create_pascal_tf_record.py文件復制到dataset文件夾下,做如下修改：
（1）修改第85行：img_path = os.path.join(data['folder'], image_subdirectory, data['filename'])
改為：img_path = os.path.join("VOC2007", image_subdirectory, data['filename'])
原因：因為我的數據標注的xml文件中的folder項是"hand_2",但是我本地並沒有該目錄，所以直接改為"VOC2007"。
（2）修改第163行：examples_path = os.path.join(data_dir, year, 'ImageSets', 'Main','aeroplane_' + FLAGS.set + '.txt')
改為：examples_path = os.path.join(data_dir, year, 'ImageSets', 'Main',FLAGS.set + '.txt')
原因：我的Main中的txt文件中沒有aeroplane_前綴
（3）根據自己的標簽創建pascal_label_map.pbtxt 文件，內容如下：

（4）運行以下命令，就可以得到用於訓練和驗證的tf_record文件：
python create_pascal_tf_record.py
--data_dir=./VOCdevkit
--label_map_path=./pascal_label_map.pbtxt
--year=VOC2007
--set=train
--output_path=./pascal_train.record

python create_pascal_tf_record.py
--data_dir=./VOCdevkit
--label_map_path=./pascal_label_map.pbtxt
--year=VOC2007
--set=val
--output_path=./pascal_val.record
此處寫的是相對路徑，若有需要可改為絕對路徑。 運行完成后將會在 目錄下得到pascal_train.record和pascal_val.record兩個文件，訓練集和驗證集的二進制文件。

3.解壓SSDMobilenet模型

tar -xvf ssd_mobilenet_v1_coco_2018_01_28.tar得到如下文件：


將文件夾里面的model.ckpt.*的三個文件copy到dataset文件夾。

4.修改config文件

將文件models/research/object_detection/samples/configs/ssd_mobilenet_v1_pets.config復制到dataset.修改：
（1）num_classes修改為自己的類別數目，我的是6
（2）修改路徑。（5處）
　　fine_tune_checkpoint: "./models/dataset/model.ckpt"
　　input_path: "./models/dataset/pascal_train.record"
　　label_map_path: ".models/dataset/pascal_label_map.pbtxt"
　　input_path: "./models/dataset/pascal_val.record"
　　label_map_path: "./models/dataset/pascal_label_map.pbtxt"
此處建議寫為絕對路徑
保存config文件，重命名為ssd_mobilenet_v1_pascal.config。

5.開始訓練

將models/research/object_detection/model_main.py文件復制到dataset路徑下：
在model_main.py文件中加三行代碼：

import os
os.environ['CUDA_VISIBLE_DEVICES'] = '2'     # 修改為當前能用的GPU
tf.logging.set_verbosity(tf.logging.INFO)     # 打印日志

執行訓練命令
python model_main.py
--./ssd_mobilenet_v1_pascal.config
--model_dir=./output
--num_train_steps=50000
--sample_1_of_n_eval_examples=1 --alsologtostderr

6.評估模型

暫時還沒找到評估模型的文件

7.查看結果

tensorboard --logdir=./models/dataset/output --port=6006
可以在瀏覽器打開http://服務器IP:6005/ 頁面觀察訓練過程

 

主要是觀察loss和mAP@.50IOU

8.生成可以被調用的模型

將models/research/object_detection目錄下的export_inference_graph.py文件復制到dataset路徑下
python export_inference_graph.py
--input_type=image_tensor
--pipeline_config_path=./ssd_mobilenet_v1_pascal.config
--trained_checkpoint_prefix=./output/model.ckpt-10000
--output_directory=./savedModelcd

生成的模型如圖所示：

9.調用生成的模型

在dataset目錄下創建object_detection_test.py,並將其復制到models/research/object_detection目錄下，因為要調用該目錄下的utils.py文件
可以在dataset下創建你自己的測試文件夾，然后更改object_detection_test.py的相應的路徑

import numpy as np
import os
import six.moves.urllib as urllib
import sys
import tarfile
import tensorflow as tf
import zipfile

from collections import defaultdict
from io import StringIO
from matplotlib import pyplot as plt
from PIL import Image

## This is needed to display the images.
#%matplotlib inline

# This is needed since the notebook is stored in the object_detection folder.
sys.path.append("..")

from utils import label_map_util

from utils import visualization_utils as vis_util

# What model to download.
#MODEL_NAME = 'ssd_mobilenet_v1_coco_2017_11_17'
#MODEL_FILE = MODEL_NAME + '.tar.gz'
#DOWNLOAD_BASE = #'http://download.tensorflow.org/models/object_detection/'
MODEL_NAME = '/home/minelab/chenqingyun/models/dataset/savedModelcd'

# Path to frozen detection graph. This is the actual model that is used for the object detection.
PATH_TO_CKPT = MODEL_NAME + '/frozen_inference_graph.pb'

# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = "/home/minelab/chenqingyun/models/dataset/pascal_label_map.pbtxt"

NUM_CLASSES = 6

#download model
#opener = urllib.request.URLopener()
#opener.retrieve(DOWNLOAD_BASE + MODEL_FILE, MODEL_FILE)
#tar_file = tarfile.open(MODEL_FILE)
#for file in tar_file.getmembers():
#  file_name = os.path.basename(file.name)
#  if 'frozen_inference_graph.pb' in file_name:
#    tar_file.extract(file, os.getcwd())

#Load a (frozen) Tensorflow model into memory.
detection_graph = tf.Graph()
with detection_graph.as_default():
  od_graph_def = tf.GraphDef()
  with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
    serialized_graph = fid.read()
    od_graph_def.ParseFromString(serialized_graph)
    tf.import_graph_def(od_graph_def, name='')
#Loading label map
label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
category_index = label_map_util.create_category_index(categories)
#Helper code
def load_image_into_numpy_array(image):
  (im_width, im_height) = image.size
  return np.array(image.getdata()).reshape(
      (im_height, im_width, 3)).astype(np.uint8)

# For the sake of simplicity we will use only 2 images:
# image1.jpg
# image2.jpg
# If you want to test the code with your images, just add path to the images to the TEST_IMAGE_PATHS.
PATH_TO_TEST_IMAGES_DIR = '/home/minelab/chenqingyun/models/dataset/1'
#TEST_IMAGE_PATHS = [ os.path.join(PATH_TO_TEST_IMAGES_DIR, 'image{}.jpg'.format(i)) for i in range(1, 3) ]
TEST_IMAGE = "2.jpeg"
#print('the test image is:', TEST_IMAGE)
TEST_IMAGE = os.path.join(PATH_TO_TEST_IMAGES_DIR,TEST_IMAGE)
# Size, in inches, of the output images.
IMAGE_SIZE = (224, 224)
IMGES_LIST = os.listdir(PATH_TO_TEST_IMAGES_DIR)
with detection_graph.as_default():
    with tf.Session(graph=detection_graph) as sess:
        for IMG_NAME in IMGES_LIST:
            TEST_IMAGE = os.path.join(PATH_TO_TEST_IMAGES_DIR,IMG_NAME)
            # Definite input and output Tensors for detection_graph
            image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
            # Each box represents a part of the image where a particular object was detected.
            detection_boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
            # Each score represent how level of confidence for each of the objects.
            # Score is shown on the result image, together with the class label.
            detection_scores = detection_graph.get_tensor_by_name('detection_scores:0')
            detection_classes = detection_graph.get_tensor_by_name('detection_classes:0')
            num_detections = detection_graph.get_tensor_by_name('num_detections:0')
            #for image_path in TEST_IMAGE_PATHS:
            image = Image.open(TEST_IMAGE)
            # the array based representation of the image will be used later in order to prepare the
            # result image with boxes and labels on it.
            image_np = load_image_into_numpy_array(image)
            # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
            image_np_expanded = np.expand_dims(image_np, axis=0)
            image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
            # Each box represents a part of the image where a particular object was detected.
            boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
            # Each score represent how level of confidence for each of the objects.
            # Score is shown on the result image, together with the class label.
            scores = detection_graph.get_tensor_by_name('detection_scores:0')
            classes = detection_graph.get_tensor_by_name('detection_classes:0')
            num_detections = detection_graph.get_tensor_by_name('num_detections:0')
            # Actual detection.
            (boxes, scores, classes, num_detections) = sess.run(
                [boxes, scores, classes, num_detections],
                feed_dict={image_tensor: image_np_expanded})
            # Visualization of the results of a detection.
            vis_util.visualize_boxes_and_labels_on_image_array(
                image_np,
                np.squeeze(boxes),
                np.squeeze(classes).astype(np.int32),
                np.squeeze(scores),
                category_index,
                use_normalized_coordinates=True,
                line_thickness=8)

            # print(scores)  
            # print(classes)  
            # print(category_index) 
           
            final_score = np.squeeze(scores)    
            count = 0
            for i in range(100):
                if scores is None or final_score[i] > 0.5:
                    count = count + 1
                    print(IMG_NAME,classes[0][i],scores[0][i])
            #print ('the count of objects is: ', count)    

            # plt.figure(figsize=IMAGE_SIZE)
            # plt.imshow(image_np)
            # plt.show()