【天天向上】OpenVINO学习笔记（四）目标检测模型

import numpy as np
import tensorflow as tf
import cv2 as cv

# Read the graph.
with tf.gfile.FastGFile('frozen_inference_graph.pb', 'rb') as f:
    graph_def = tf.GraphDef()
    graph_def.ParseFromString(f.read())

with tf.Session() as sess:
    # Restore session
    sess.graph.as_default()
    tf.import_graph_def(graph_def, name='')

    # Read and preprocess an image.
    img = cv.imread('example.jpg')
    rows = img.shape[0]
    cols = img.shape[1]
    inp = cv.resize(img, (300, 300))
    inp = inp[:, :, [2, 1, 0]]  # BGR2RGB

    # Run the model
    out = sess.run([sess.graph.get_tensor_by_name('num_detections:0'),
                    sess.graph.get_tensor_by_name('detection_scores:0'),
                    sess.graph.get_tensor_by_name('detection_boxes:0'),
                    sess.graph.get_tensor_by_name('detection_classes:0')],
                   feed_dict={'image_tensor:0': inp.reshape(1, inp.shape[0], inp.shape[1], 3)})

    # Visualize detected bounding boxes.
    num_detections = int(out[0][0])
    for i in range(num_detections):
        classId = int(out[3][0])
        score = float(out[1][0])
        bbox = [float(v) for v in out[2][0]]
        if score > 0.3:
            x = bbox[1] * cols
            y = bbox[0] * rows
            right = bbox[3] * cols
            bottom = bbox[2] * rows
            cv.rectangle(img, (int(x), int(y)), (int(right), int(bottom)), (125, 255, 51), thickness=2)

cv.imshow('TensorFlow MobileNet-SSD', img)
cv.waitKey()
复制代码

【opencv】
如果电脑没有tensorflow想要恢复模型怎么办？那就可以考虑下opencv了，opencv加载tensorflow模型需要pb文件和pbtxt文件，pbtxt是可以根据pb文件生成的，在opencv的源代码中进入sample/dnn 文件夹中，由于我们的模型是ssd因此找到以下两个文件

tf_text_graph_ssd.py
tf_text_graph_common.py
在https://github.com/opencv/opencv，中的 opencv-master\samples\dnn中找到

在命令行执行

python tf_text_graph_ssd.py --input C:\Users\zlzx\Downloads\ssd_mobilenet_v2_coco_2018_03_29\frozen_inference_graph.pb --config C:\Users\zlzx\Downloads\ssd_mobilenet_v2_coco_2018_03_29\pipeline.config --output frozen_inference_graph.pbtxt
frozen_inference_graph.pb是下载的训练pb模型，pipeline是训练时的配置文件，生成pb时会有，frozen_inference_graph.pbtxt就是生成的文件用于opencv恢复模型

然后用opencv恢复模型

大致代码如下：

import cv2 as cv

cvNet = cv.dnn.readNetFromTensorflow('frozen_inference_graph.pb', 'frozen_inference_graph.pbtxt')

img = cv.imread('example.jpg')
rows = img.shape[0]
cols = img.shape[1]
cvNet.setInput(cv.dnn.blobFromImage(img, size=(300, 300), swapRB=True, crop=False))
cvOut = cvNet.forward()

for detection in cvOut[0,0,:,:]:
    score = float(detection[2])
    if score > 0.3:
        left = detection[3] * cols
        top = detection[4] * rows
        right = detection[5] * cols
        bottom = detection[6] * rows
        cv.rectangle(img, (int(left), int(top)), (int(right), int(bottom)), (23, 230, 210), thickness=2)

cv.imshow('img', img)
cv.waitKey()
复制代码