tensorflow，object，detection，在model zoom，新下載的模型，WARNING:root:Variable [resnet_v1_50/block1/unit_3/bot

現象：

WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/BatchNorm/beta] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/BatchNorm/beta/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/BatchNorm/gamma] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/BatchNorm/gamma/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/BatchNorm/moving_mean] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/BatchNorm/moving_variance] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/weights] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/weights/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv2/BatchNorm/beta] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv2/BatchNorm/beta/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv2/BatchNorm/gamma] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv2/BatchNorm/gamma/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv2/BatchNorm/moving_mean] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv2/BatchNorm/moving_variance] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv2/weights] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv2/weights/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv3/BatchNorm/beta] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv3/BatchNorm/beta/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv3/BatchNorm/gamma] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv3/BatchNorm/gamma/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv3/BatchNorm/moving_mean] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv3/BatchNorm/moving_variance] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv3/weights] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv3/weights/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/shortcut/BatchNorm/beta] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/shortcut/BatchNorm/beta/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/shortcut/BatchNorm/gamma] is not available in checkpoint

 

這個報的warning是你要的變量在 checkpoint裏面找不到！！！！！！！！！！！

那checkpoint裏面是什麼？？？？？？？？？？？？？那就去看咯：

而後就吧checkpoint的信息打印出來：

import tensorflow as tf
from tensorflow.python.tools.inspect_checkpoint import print_tensors_in_checkpoint_file


latest_ckp = tf.train.latest_checkpoint('./')  # checkpoint 所在的路徑是當前文件夾（./）,若是是其餘，請改過來
print_tensors_in_checkpoint_file(latest_ckp, all_tensors=True, tensor_name='')
而後發現：

tensor_name:  FeatureExtractor/resnet_v1_50/block4/unit_2/bottleneck_v1/conv3/BatchNorm/beta
[-1.2371869 -1.3236059 -1.2312554 ... -1.3017322 -1.2117504 -1.4911506]
tensor_name:  FeatureExtractor/resnet_v1_50/block4/unit_2/bottleneck_v1/conv3/BatchNorm/gamma
[1.4568403 1.4048293 1.0624585 ... 1.1037463 1.0324904 1.0967498]
tensor_name:  FeatureExtractor/resnet_v1_50/block4/unit_2/bottleneck_v1/conv3/BatchNorm/moving_mean
[-0.08689928  0.00798311 -0.02823892 ... -0.04347016 -0.07710622
 -0.04719209]
tensor_name:  FeatureExtractor/resnet_v1_50/block4/unit_2/bottleneck_v1/conv3/BatchNorm/moving_variance
[0.00452107 0.00487971 0.00305854 ... 0.00261809 0.00344088 0.00306313]
tensor_name:  FeatureExtractor/resnet_v1_50/block4/unit_2/bottleneck_v1/conv3/weights
[[[[ 1.2295754e-03 -9.4967345e-03 -6.6470391e-05 ... -9.7579239e-03
     1.1851139e-02 -1.0630138e-02]
   [-4.5038795e-04 -3.9205588e-03 -6.4933673e-03 ...  9.1390898e-03
    -1.1232623e-02 -9.8358802e-03]
   [ 1.3918030e-02 -7.1829297e-03  3.0942420e-03 ... -6.6203251e-03

原來是少了FeatureExtractor，這時候就去找restore var 的code
var_name = (
 re.split('^' + self._extract_features_scope + '/',
 var_name)[-1])FeatureExtractor

他把這個給幹掉了，因此：
問題是，tensorflow object detection 在model zoom新下載的模型命名方式改變了，若是使用舊的代碼加載新的模型就會出現這個問題，解決方案是research/object_detection/meta_architectures/ssd_meta_arch.py_extract_features_scope就是：FeatureExtractor
發現_extract_features_scopeFeatureExtractor

var_name = (
    re.split('^' + self._extract_features_scope + '/',
 var_name)[-1])
改成：

var_name = (
    re.split('^' + '/',
 var_name)[-1])

一樣的其餘模型也是這樣。

可是，後面仍是出現：

WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/BatchNorm/beta/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/BatchNorm/gamma/Momentum] is not available in checkpoint
WARNING:root:Variable [resnet_v1_50/block1/unit_1/bottleneck_v1/conv1/weights/Momentum] is not available in checkpoint

仔細看，你會發現，這只是Momentum形成的，是優化器中的參數，不是模型參數，Momentum是動量的意思，優化器使用這個是爲了

避免收斂到局部極值。這個不影響的。若是不想報錯能夠：

if fine_tune_checkpoint_type == 'classification':
 var_name = (
 re.split('^' + '/',
 var_name)[-1])
 # var_name = (
 # re.split('^' + self._extract_features_scope + '/',
 # var_name)[-1])
if 'Momentum' in var_name:
 continue
variables_to_restore[var_name] = variable

加一個：

if 'Momentum' in var_name:
 continue

若是是Moment 變量就不加載。

不過，若是對這套代碼不瞭解的，建議update最新的代碼。