Search code examples
pythoncomputer-visiontensorflow2.0tensorflow-lite

Convert .pb file to .tflite

I trained a custom dataset model using TensorFlow object detection API. Tensorflow version =2.2.0. I was trying to convert the .pb file to .tflite using the following code and getting the following error:

import tensorflow as tf

# Convert the model
converter = tf.lite.TFLiteConverter.from_saved_model('/content/drive/MyDrive/FINAL DNET MODEL/inference_graph/saved_model') # path to the SavedModel directory
tflite_model = converter.convert()

# Save the model.
with open('model.tflite', 'wb') as f:
  
  f.write(tflite_model)

Error message:

--------------------------------------------------------------------------- Exception                                 Traceback (most recent call last) /usr/local/lib/python3.6/dist-packages/tensorflow/lite/python/convert.py in toco_convert_protos(model_flags_str, toco_flags_str, input_data_str, debug_info_str, enable_mlir_converter)
    212                                                  debug_info_str,
--> 213                                                  enable_mlir_converter)
    214       return model_str

4 frames Exception: <unknown>:0: error: loc(callsite(callsite("map/TensorArrayV2_1@__inference_call_func_18902" at "StatefulPartitionedCall@__inference_signature_wrapper_23056") at "StatefulPartitionedCall")): requires element_shape to be 1D tensor during TF Lite transformation pass <unknown>:0: note: loc("StatefulPartitionedCall"): called from <unknown>:0: error: loc(callsite(callsite("map/TensorArrayV2_1@__inference_call_func_18902" at "StatefulPartitionedCall@__inference_signature_wrapper_23056") at "StatefulPartitionedCall")): failed to legalize operation 'tf.TensorListReserve' that was explicitly marked illegal <unknown>:0: note: loc("StatefulPartitionedCall"): called from


During handling of the above exception, another exception occurred:

ConverterError                            Traceback (most recent call last) /usr/local/lib/python3.6/dist-packages/tensorflow/lite/python/convert.py in toco_convert_protos(model_flags_str, toco_flags_str, input_data_str, debug_info_str, enable_mlir_converter)
    214       return model_str
    215     except Exception as e:
--> 216       raise ConverterError(str(e))
    217 
    218   if distutils.spawn.find_executable(_toco_from_proto_bin) is None:

ConverterError: <unknown>:0: error: loc(callsite(callsite("map/TensorArrayV2_1@__inference_call_func_18902" at "StatefulPartitionedCall@__inference_signature_wrapper_23056") at "StatefulPartitionedCall")): requires element_shape to be 1D tensor during TF Lite transformation pass <unknown>:0: note: loc("StatefulPartitionedCall"): called from <unknown>:0: error: loc(callsite(callsite("map/TensorArrayV2_1@__inference_call_func_18902" at "StatefulPartitionedCall@__inference_signature_wrapper_23056") at "StatefulPartitionedCall")): failed to legalize operation 'tf.TensorListReserve' that was explicitly marked illegal <unknown>:0: note: loc("StatefulPartitionedCall"): called from

Please help me resolve this error and convert the .pb file to .tflite.

Solution

  • This Github issue had a similar error message. This answer suggests to specify fn_output_signature=tf.TensorSpec(shape, dtype) when calling map function.