detect.tflite 将 TF model 转换为 TFlite 失败

Question

I am using my mac, running Catalina 10.15.5, with Python 2.7.17, running Tensorflow 1.15.3 from source, installed using pip, not in Anaconda, and my bazel version is 0.26.1.

我正在尝试按照本教程将 ssd_mobilenet_v3_small_coco 上的自定义 model 转换为 TFlite： 在步骤 3但无法使用此行将我的 tflite_graph.pb 转换为 detect.tflite：

bazel run --config=opt tensorflow/lite/toco:toco -- --input_file=/Users/jp3spinelli/Desktop/models/research/object_detection/TFLite_model/tflite_graph.pb --output_file=/Users/jp3spinelli/Desktop/models/research/object_detection/TFLite_model/detect.tflite --input_shapes=1,300,300,3 --input_arrays=normalized_input_image_tensor --output_arrays=TFLite_Detection_PostProcess,TFLite_Detection_PostProcess:1,TFLite_Detection_PostProcess:2,TFLite_Detection_PostProcess:3 --inference_type=FLOAT --allow_custom_ops

我一直遇到这个错误（我只包括最后一点，因为它很长）：

WARNING: /Users/jp3spinelli/tensorflow/tensorflow/core/BUILD:2455:12: in srcs attribute of cc_library rule //tensorflow/core:lib_internal_impl: please do not import '//tensorflow/core/lib/strings:proto_text_util.cc' directly. You should either move the file to this package or depend on an appropriate rule there
WARNING: /Users/jp3spinelli/tensorflow/tensorflow/core/BUILD:2455:12: in srcs attribute of cc_library rule //tensorflow/core:lib_internal_impl: please do not import '//tensorflow/core/lib/strings:scanner.cc' directly. You should either move the file to this package or depend on an appropriate rule there
WARNING: /Users/jp3spinelli/tensorflow/tensorflow/core/BUILD:2455:12: in srcs attribute of cc_library rule //tensorflow/core:lib_internal_impl: please do not import '//tensorflow/core/lib/strings:strcat.cc' directly. You should either move the file to this package or depend on an appropriate rule there
INFO: Analyzed target //tensorflow/lite/toco:toco (0 packages loaded, 0 targets configured).
INFO: Found 1 target...
Target //tensorflow/lite/toco:toco up-to-date:
  bazel-bin/tensorflow/lite/toco/toco
INFO: Elapsed time: 0.472s, Critical Path: 0.00s
INFO: 0 processes.
INFO: Build completed successfully, 1 total action
INFO: Running command line: bazel-bin/tensorflow/lite/toco/toco '--input_file=/Users/jp3spinelli/Desktop/models/research/object_detection/TFLite_model/tflite_graph.pb' '--output_file=/Users/jp3spinelli/Desktop/models/research/object_detection/TFLite_model/detect.tflite' '--input_shapes=1,300,300,3' '--input_arrays=normalized_input_image_tensor' '--output_arrays=TFLite_Detection_PostProcess,TFLite_Detection_PostProcess:1,TFLite_Detection_PostProcess:2,TFLite_DetectioINFO: Build completed successfully, 1 total action
2020-07-21 18:03:47.675656: I tensorflow/lite/toco/import_tensorflow.cc:659] Converting unsupported operation: TFLite_Detection_PostProcess
2020-07-21 18:03:47.691874: F tensorflow/lite/toco/tooling_util.cc:1669] Check failed: input_array_dims[i] == input_array_proto.shape().dims(i) (320 vs. 300)
Abort trap: 6

它正在创建一个文件，但它的字节数为零，所以我知道出了点问题。 我正在使用我的主目录中的 tensorflow 目录。 我注意到的一件奇怪的事情是，我桌面上“模型”文件夹的路径显示它从“iCloud Drive”而不是“用户”开始。

请让我知道如何解决这个问题，我是编码新手，所以我需要一些分步说明。 谢谢！

Answer 1

我认为您应该将 '' 之间的 output_arrays 设置为字符串，尝试运行如下内容：

bazel run --config=opt tensorflow/lite/toco:toco -- \
--input_file=$OUTPUT_DIR/tflite_graph.pb \
--output_file=$OUTPUT_DIR/detect.tflite \
--input_shapes=1,300,300,3 \
--input_arrays=normalized_input_image_tensor \
--output_arrays='TFLite_Detection_PostProcess','TFLite_Detection_PostProcess:1','TFLite_Detection_PostProcess:2','TFLite_Detection_PostProcess:3'  \
--inference_type=FLOAT \
--allow_custom_ops