1#!/usr/bin/python
2# -*- coding: UTF-8 -*- 
3
4import errno
5import math
6import os
7import sys
8import json
9
10import cv2
11import numpy as np
12
13import paddlemobile as pm
14
15predictor = None
16labels = []
17
18def init(configs):
19    global predictor
20
21    model_dir = configs['model']
22    pm_config = pm.PaddleMobileConfig()   # 创建paddlemobile配置对象
23    pm_config.precision = pm.PaddleMobileConfig.Precision.FP32  # 对精度赋值成FP32
24    pm_config.device = pm.PaddleMobileConfig.Device.kFPGA  # 选定FPGA
25    pm_config.prog_file = model_dir + "/model"
26    pm_config.param_file = model_dir + '/params'
27    pm_config.thread_num = 4
28
29    print('')
30    print('configuration for predictor is :')
31    print('\tPrecision: ' + str(pm_config.precision))
32    print('\t   Device: ' + str(pm_config.device))
33    print('\t    Model: ' + str(pm_config.prog_file))
34    print('\t   Params: ' + str(pm_config.param_file))
35    print('\tThreadNum: ' + str(pm_config.thread_num))
36    print('')
37
38    predictor = pm.CreatePaddlePredictor(pm_config)  # 创建预测器
39
40# 读取label_list.txt文件
41def read_labels(configs):
42    global labels
43    if not 'labels' in configs:
44        return
45    label_path = configs['labels']
46    if label_path is None or label_path == '':
47        return
48    with open(label_path) as label_file:
49        line = label_file.readline()
50        while line:
51            labels.append(line.strip().split(':')[-1])
52            line = label_file.readline()
53
54# 读取本地图片            
55def read_image(configs):
56    image_input = cv2.imread(configs['image'], cv2.IMREAD_COLOR)
57    return image_input
58
59# 图像预处理
60def preprocess_image(image_input, configs):
61    # resizing image
62    print('image shape input: ' + str(image_input.shape))
63    width = configs['input_width']
64    height = configs['input_height']
65    image_resized = cv2.resize(image_input, (width, height), cv2.INTER_CUBIC)
66    print('image shape resized: ' + str(image_resized.shape))
67
68    # to float32
69    image = image_resized.astype('float32')
70
71    # transpose to channel-first format
72    image_transposed = np.transpose(image, (2, 0, 1))
73    print('image shape transposed: ' + str(image_transposed.shape))
74
75    # mean and scale preprocessing
76    mean  = np.array(configs['mean']).reshape((3, 1, 1))
77    scale_number = configs['scale']
78    scale = np.array([scale_number, scale_number, scale_number]).reshape((3, 1, 1))
79
80    # RGB or BGR formatting
81    format = configs['format'].upper()
82    if format == 'RGB':
83        b = image_transposed[0]
84        g = image_transposed[1]
85        r = image_transposed[2]
86        image_transposed = np.stack([r, g, b])
87        print('image shape formatted transposed: ' + str(image_transposed.shape))
88
89    # mean and scale
90    print 'substract mean', mean.flatten(), ' and multiple with scale', scale.flatten()
91    image_transposed -= mean
92    image_transposed *= scale
93
94    # transpose back
95    image_result = np.transpose(image_transposed, (1, 2, 0))
96    print('image shape transposed-back: ' + str(image_result.shape))
97
98    print('')
99    return image_result
100
101# 检测模型输出结果图片
102def draw_results(image, output, threshold):
103    height, width, _ = image.shape
104    print 'boxes with scores above the threshold (%f): \n' % threshold
105    i = 1
106    for box in output:
107        if box[1] > threshold:
108            print '\t', i, '\t', box[0], '\t', box[1], '\t', box[2], '\t', box[3], '\t', box[4], '\t', box[5]
109            x_min = int(box[2]  * width )
110            y_min = int(box[3]  * height)
111            x_max = int(box[4]  * width )
112            y_max = int(box[5]  * height)
113            cv2.rectangle(image, (x_min, y_min), (x_max, y_max), (0, 255, 0), 3)
114            i += 1
115    cv2.imwrite('result.jpg', image)
116    print('')
117
118# 分类模型预测结果输出
119def classify(output, configs):
120    data = output.flatten()
121    max_index = 0
122    score = 0.0
123    for i in range(len(data)):
124        if data[i] > score and not data[i] == float('inf'):
125            max_index = i
126            score = data[i]
127
128    print 'label: ', labels[max_index]
129    print 'index: ', max_index
130    print 'score: ', score
131    print ''
132
133# 检测模型预测结果输出    
134def detect(output, configs):
135    image = read_image(configs)
136    draw_results(image, output, configs['threshold'])
137
138# 模型预测
139def predict(configs, detection):
140    global predictor
141    width = configs['input_width']
142    height = configs['input_height']
143
144    image = read_image(configs)
145    input = preprocess_image(image, configs)
146
147    tensor = pm.PaddleTensor()
148    tensor.dtype = pm.PaddleDType.FLOAT32
149    tensor.shape = (1, 3, width, height)
150    tensor.data = pm.PaddleBuf(input)
151
152    paddle_data_feeds = [tensor]
153
154    print('prediction is running ...')
155    outputs = predictor.Run(paddle_data_feeds)
156    assert len(outputs) == 1, 'error numbers of tensor returned from Predictor.Run function !!!'
157
158    output = np.array(outputs[0], copy = False)
159
160    print('\nprediction result :')
161    print('\t nDim: ' + str(output.ndim))
162    print('\tShape: ' + str(output.shape))
163    print('\tDType: ' + str(output.dtype))
164    print('')
165    # print(output)
166    # print('')
167    if detection:
168        detect(output, configs)
169    else:
170        classify(output, configs)

1#!/usr/bin/python
2# -*- coding: UTF-8 -*- 
3
4import errno
5import math
6import os
7import sys
8import argparse
9import json
10
11import cv2
12import numpy as np
13
14import edgeboard
15
16# 参数设定
17def parse_args():
18    parser = argparse.ArgumentParser(description='API implementation for Paddle-Mobile')
19    parser.add_argument('-d', '--detection',
20                        help='flag indicating detections',
21                        action="store_true")
22    parser.add_argument('-j', '--json',
23                        help='configuration file for the prediction')
24    return parser.parse_args()
25
26# 输出参数
27def print_args(args):
28    print 'Arguments: '
29    print '\t', '    detection flag: ', args.detection
30    print '\t', 'json configuration: ', args.json
31
32# 主函数
33def main():
34    args = parse_args()
35    print_args(args)
36    if args.json is None or args.json == '':
37        print '\nFor usage, please use the -h switch.\n\n'
38        sys.exit(0)
39
40    with open(args.json) as json_file:
41        configs = json.load(json_file)
42
43    edgeboard.init(configs)
44    edgeboard.read_labels(configs)
45    edgeboard.predict(configs, args.detection)
46
47if __name__ == '__main__':
48    sys.exit(main())

1//进入系统根目录
2cd /
3
4//展开tar.gz 压缩包
5tar -xzvf  home/root/workspace/paddlemobile-0.0.1.linux-aarch64-py2.tar.gz
6
7//返回用户HOME目录
8cd
9
10//检查package paddlemobile (0.0.1) 是否已经安装
11pip list

1 python api.py -j {分类模型json文件}

1 python api.py -j configs.classification/Inceptionv3/zebra.json

1python api.py -d -j {检测模型json文件}

1python api.py -d -j configs.detection/vgg-ssd/screw.json