OpenVINO不能转换Mixer模型

Question

我使用这个GitHub存储库来训练Mixer 2.5.0模型。

​

​

我尝试使用以下命令生成.bin和.xml文件

mo --data_type FP16 --saved_model_dir C:\Users\john0\Desktop\mlp --input_shape (1,150,150,3)

下面是我面临的错误。

[ WARNING ]  Failed to parse a tensor with Unicode characters. Note that Inference Engine does not support string literals, so the string constant should be eliminated from the graph.
[ WARNING ]  Failed to parse a tensor with Unicode characters. Note that Inference Engine does not support string literals, so the string constant should be eliminated from the graph.
[ ERROR ]  Exception occurred during running replacer "REPLACEMENT_ID" (<class 'openvino.tools.mo.front.user_data_repack.UserDataRepack'>): Original placeholders: 'serving_default_input_1, saver_filename'. Freezing was requested for ''. --input_shape was provided without --input. Can not deduce which node shape to override

我使用openvino_2022.1.0.643版本。

你可以下载我的模型这里。

Answer 1

错误是由于模型具有多个输入，并且可以使用这个MO命令mo --data_type FP16 --saved_model_dir model\directory\mlp\ --input_shape (1..,150,150,3)来解决。然而，我现在收到了不同的错误：

[ ERROR ]  List of operations that cannot be converted to Inference Engine IR:
[ ERROR ]      FusedBatchNormV3 (16)
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block/layer_normalization/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block/layer_normalization_1/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_1/layer_normalization_2/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_1/layer_normalization_3/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_2/layer_normalization_4/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_2/layer_normalization_5/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_3/layer_normalization_6/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_3/layer_normalization_7/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_4/layer_normalization_8/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_4/layer_normalization_9/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_5/layer_normalization_10/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_5/layer_normalization_11/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_6/layer_normalization_12/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_6/layer_normalization_13/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_7/layer_normalization_14/FusedBatchNormV3
[ ERROR ]          StatefulPartitionedCall/mlp_mixer/mlp_block_7/layer_normalization_15/FusedBatchNormV3
[ ERROR ]  Part of the nodes was not converted to IR. Stopped.

如您所见，您的模型中的FusedBatchNormV3层在TensorFlow 2中不受支持，您可以参考支持框架层获取支持的TensorFlow 2操作列表。