添付のコードを、Edge TPU PythonAPI 仕様から、PyCoralAPI（python)仕様に作り直したい。

#■やりたいこと
Coral devboard mendel（Linux）環境でSmart Bird Feeder project.を試したいと思っています。（https://coral.ai/projects/bird-feeder/）

添付コード bird_classify.pyを実行しようとしていますが実行できず、次のようなエラーになります。

#■エラー

error内容
1engine = ClassificationEngine(args.model)
2NameError: name 'ClassificationEngine' is not defined

pycoral apiを使って pycoralを使うように書き換え、コードを作り直す必要があることまではわかるのですが、具体的にどのように直すと良いでしょうか。良い方法があれば教えていただければ幸いです。

#■やってみたこと
ClassificationEngineの箇所をpycoral仕様にするため、edgetpu.classification.engineを削除し、pycoralで必要になりそうなコードを追加しようと思い、次のように記述しました。

bird_classify.py　もともとあった記述（削除済み）
1from edgetpu.classification.engine import ClassificationEngine

https://coral.ai/docs/reference/py/pycoral.utils/#module-pycoral.utils.dataset　を参照し、ClassificationEngineの代わりになりそうなもの？と思い次のようにしてみました。

bird_classify.py　追加
1from pycoral.adapters import classify
2from pycoral.adapters import common
3from pycoral.utils.dataset import read_label_file
4from pycoral.utils.edgetpu import make_interpreter

bird_classify.py　全文　python3
1import argparse
2import time
3import re
4import imp
5import logging
6import gstreamer
7
8from PIL import Image
9from playsound import playsound
10
11from pycoral.adapters import classify
12from pycoral.adapters import common
13from pycoral.utils.dataset import read_label_file
14from pycoral.utils.edgetpu import make_interpreter
15
16def save_data(image,results,path,ext='png'):
17    """Saves camera frame and model inference results
18    to user-defined storage directory."""
19    tag = '%010d' % int(time.monotonic()*1000)
20    name = '%s/img-%s.%s' %(path,tag,ext)
21    image.save(name)
22    print('Frame saved as: %s' %name)
23    logging.info('Image: %s Results: %s', tag,results)
24
25def load_labels(path):
26    """Parses provided label file for use in model inference."""
27    p = re.compile(r'\s*(\d+)(.+)')
28    with open(path, 'r', encoding='utf-8') as f:
29      lines = (p.match(line).groups() for line in f.readlines())
30      return {int(num): text.strip() for num, text in lines}
31
32def print_results(start_time, last_time, end_time, results):
33    """Print results to terminal for debugging."""
34    inference_rate = ((end_time - start_time) * 1000)
35    fps = (1.0/(end_time - last_time))
36    print('\nInference: %.2f ms, FPS: %.2f fps' % (inference_rate, fps))
37    for label, score in results:
38      print(' %s, score=%.2f' %(label, score))
39
40def do_training(results,last_results,top_k):
41    """Compares current model results to previous results and returns
42    true if at least one label difference is detected. Used to collect
43    images for training a custom model."""
44    new_labels = [label[0] for label in results]
45    old_labels = [label[0] for label in last_results]
46    shared_labels  = set(new_labels).intersection(old_labels)
47    if len(shared_labels) < top_k:
48      print('Difference detected')
49      return True
50
51def user_selections():
52    parser = argparse.ArgumentParser()
53    parser.add_argument('--model', required=True,
54                        help='.tflite model path')
55    parser.add_argument('--labels', required=True,
56                        help='label file path')
57    parser.add_argument('--top_k', type=int, default=3,
58                        help='number of classes with highest score to display')
59    parser.add_argument('--threshold', type=float, default=0.1,
60                        help='class score threshold')
61    parser.add_argument('--storage', required=True,
62                        help='File path to store images and results')
63    parser.add_argument('--sound', required=True,
64                        help='File path to deterrent sound')
65    parser.add_argument('--print', default=False, required=False,
66                        help='Print inference results to terminal')
67    parser.add_argument('--training', default=False, required=False,
68                        help='Training mode for image collection')
69    args = parser.parse_args()
70    return args
71
72
73def main():
74    
75    args = user_selections()
76    print("Loading %s with %s labels."%(args.model, args.labels))
77    engine = ClassificationEngine(args.model)
78    labels = load_labels(args.labels)
79    storage_dir = args.storage
80
81    
82    logging.basicConfig(filename='%s/results.log'%storage_dir,
83                        format='%(asctime)s-%(message)s',
84                        level=logging.DEBUG)
85
86    last_time = time.monotonic()
87    last_results = [('label', 0)]
88    def user_callback(image,svg_canvas):
89        nonlocal last_time
90        nonlocal last_results
91        start_time = time.monotonic()
92        results = engine.classify_with_image(image, threshold=args.threshold, top_k=args.top_k)
93        end_time = time.monotonic()
94        results = [(labels[i], score) for i, score in results]
95
96        if args.print:
97          print_results(start_time,last_time, end_time, results)
98
99        if args.training:
100          if do_training(results,last_results,args.top_k):
101            save_data(image,results, storage_dir)
102        else:
103          
104          if results[0][0] !='background':
105            save_data(image, storage_dir,results)
106          if 'fox squirrel, eastern fox squirrel, Sciurus niger' in results:
107            playsound(args.sound)
108            logging.info('Deterrent sounded')
109
110        last_results=results
111        last_time = end_time
112    result = gstreamer.run_pipeline(user_callback)
113
114if __name__ == '__main__':
115    main()
116