【发布时间】：2018-09-30 01:52:15
【问题描述】：
我正在自定义数据集上训练对象检测模型，并为此微调来自 TensorFlow 模型动物园的 ssd_inception_v2 预训练模型。训练脚本指向 tarball 中给出的 model.ckpt.index。
TensorFlow 返回一个 ValueError 异常（没有要保存的变量），并带有关于权重不存在的警告：
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/BatchNorm/beta/RMSProp_1] is not available in checkpoint [0/1875]
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/BatchNorm/gamma] is not available in checkpoint
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/BatchNorm/gamma/ExponentialMovingAverage] is not available in checkpoint
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/BatchNorm/gamma/RMSProp] is not available in checkpoint
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/BatchNorm/gamma/RMSProp_1] is not available in checkpoint
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/BatchNorm/moving_mean] is not available in checkpoint
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/BatchNorm/moving_variance] is not available in checkpoint
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/weights] is not available in checkpoint
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/weights/ExponentialMovingAverage] is not available in checkpoint
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/weights/RMSProp] is not available in checkpoint
WARNING:root:Variable [FeatureExtractor/InceptionV2/Mixed_5c_2_Conv2d_5_3x3_s2_128/weights/RMSProp_1] is not available in checkpoint
还有更多这样的错误。我也试过 ssd_mobilenet_v2 有同样的错误。我刚刚更改了配置文件中的类数以及“PATHS_TO_BE_CONFIGURED”。
在 GPU 上使用 TensorFlow 1.7。
【问题讨论】：
-
我能够通过指向 model.ckpt 而不是 model.ckpt.index 来解决这个问题 是的，文件 model.ckpt 不存在，但你需要指向它，以便所有 3文件被挑选出来。奇怪的是，这在任何地方都没有记录。
标签： python tensorflow object-detection