yoloの学習ファイル h5➞weightsに変換したい

前提・実現したいこと

画像処理の学習モデルのh5ファイルをweightsファイルに変換でエラーがでます。
下記ソース以外でh5ファイルをweightsファイルに変換する方法などがあれば
教えて頂けても助かります。

発生している問題・エラーメッセージ

File "converter_h5-2-wts.py", line 119, in <module>
    keras_loader.close()
NameError: name 'keras_loader' is not defined

該当のソースコード

python
1# Script converter_h5-2-wts.py
2# -*-coding: utf-8 -*-
3''' yolov3_keras_to_darknet.py'''
4import argparse
5import numpy
6import numpy as np
7import keras
8from keras.models import load_model
9from keras import backend as K
10def parser():
11    parser= argparse.ArgumentParser(description="Darknet\'s yolov3.cfg and yolov3.weights \converted into Keras\'s yolov3.h5!")
12    parser.add_argument('-cfg_path', help='yolov3.cfg')
13    parser.add_argument('-h5_path', help='yolov3.h5')
14    parser.add_argument('-output_path', help='yolov3.weights')
15    return parser.parse_args()
16class WeightSaver(object):
17    def __init__(self,h5_path,output_path):
18        self.model= load_model(h5_path)
19        self.layers= {weight.name:weight for weight in self.model.weights}
20        self.sess= K.get_session()
21        self.fhandle= open(output_path,'wb')
22        self._write_head()
23    def _write_head(self):
24        numpy_data= numpy.ndarray(shape=(3,),
25                      dtype='int32',
26                      buffer=np.array([0,2,0],dtype='int32') )
27        self.save(numpy_data)
28        numpy_data= numpy.ndarray(shape=(1,),
29                      dtype='int64',
30                      buffer=np.array([320000],dtype='int64'))
31        self.save(numpy_data)
32    def get_bn_layername(self,num):
33        layer_name= 'batch_normalization_{num}'.format(num=num)
34        bias= self.layers['{0}/beta:0'.format(layer_name)]
35        scale= self.layers['{0}/gamma:0'.format(layer_name)]
36        mean= self.layers['{0}/moving_mean:0'.format(layer_name)]
37        var= self.layers['{0}/moving_variance:0'.format(layer_name)]
38        bias_np= self.get_numpy(bias)
39        scale_np= self.get_numpy(scale)
40        mean_np= self.get_numpy(mean)
41        var_np= self.get_numpy(var)
42        return bias_np,scale_np,mean_np,var_np
43def get_convbias_layername(self,num):
44    layer_name= 'conv2d_{num}'.format(num=num)
45    bias= self.layers['{0}/bias:0'.format(layer_name)]
46    bias_np= self.get_numpy(bias)
47    return bias_np
48def get_conv_layername(self,num):
49    layer_name= 'conv2d_{num}'.format(num=num)
50    conv= self.layers['{0}/kernel:0'.format(layer_name)]
51    conv_np= self.get_numpy(conv)
52    return conv_np
53def get_numpy(self,layer_name):
54    numpy_data= self.sess.run(layer_name)
55    return numpy_data
56def save(self,numpy_data):
57    bytes_data= numpy_data.tobytes()
58    self.fhandle.write(bytes_data)
59    self.fhandle.flush()
60def close(self):
61    self.fhandle.close()
62class KerasParser(object):
63    def __init__(self, cfg_path, h5_path, output_path):
64        self.block_gen= self._get_block(cfg_path)
65        self.weights_saver= WeightSaver(h5_path, output_path)
66        self.count_conv= 0
67        self.count_bn= 0
68def _get_block(self,cfg_path):
69    block= {}
70    with open(cfg_path,'r', encoding='utf-8') as fr:
71        for line in fr:
72            line= line.strip()
73            if '[' in line and ']' in line:
74                if block:
75                    yield block
76                block= {}
77                block['type']= line.strip(' []')
78            elif not line or '#' in line:
79                continue
80            else:
81                key,val= line.strip().replace(' ','').split('=')
82                key,val= key.strip(), val.strip()
83                block[key]= val
84        yield block
85def close(self):
86    self.weights_saver.close()
87def conv(self, block):
88    self.count_conv += 1
89    batch_normalize= 'batch_normalize' in block
90    print('handing.. ',self.count_conv)
91    # If bn exists, process bn first, in order of bias, scale, mean, var
92    if batch_normalize:
93        bias,scale,mean,var= self.bn()
94        self.weights_saver.save(bias)
95        scale= scale.reshape(1,-1)
96        mean= mean.reshape(1,-1)
97        var= var.reshape(1,-1)
98        remain= np.concatenate([scale,mean,var],axis=0)
99        self.weights_saver.save(remain)
100    # biase
101    else:
102        conv_bias= self.weights_saver.get_convbias_layername(self.count_conv)
103        self.weights_saver.save(conv_bias)
104    # weights
105    conv_weights= self.weights_saver.get_conv_layername(self.count_conv)
106    # (height, width, in_dim, out_dim) (out_dim, in_dim, height, width)
107    conv_weights= np.transpose(conv_weights,[3,2,0,1])
108    self.weights_saver.save(conv_weights)
109def bn(self):
110    self.count_bn += 1
111    bias,scale,mean,var= self.weights_saver.get_bn_layername(self.count_bn)
112    return bias,scale,mean,var
113def main():
114    args= parser()
115    keras_loader= KerasParser(args.cfg_path, args.h5_path, args.output_path)
116    for block in keras_loader.block_gen:
117        if 'convolutional' in block['type']:
118            keras_loader.conv(block)
119keras_loader.close()
120if __name__== "__main__":
121    main()
122

試したこと

補足情報（FW/ツールのバージョンなど）

python 3.6.3

jbpb0

2021/12/22 08:22 編集

質問に掲載されてるコードは、どこから持ってきたものでしょうか？もしこれなら、インデントが間違いまくってますよ https://github.com/OmniXRI/OpenVINO_RealSense_HarvestBot/blob/master/my_yolo3/yolov3_keras_to_darknet.py