search

lhelontra / tensorflow-on-arm

TensorFlow for Arm
MIT License
1.05k stars 274 forks source link

ImportError, undefined symbol #62

Closed Flipajs closed 5 years ago

Flipajs commented 5 years ago

Hi Leonardo, first of all, great job.

I get following error when using tensorflow-1.14.0-cp35-none-linux_armv7l.whl release

ImportError: /home/filip/.local/lib/python3.5/site-packages/tensorflow/lite/python/interpreter_wrapper/_tensorflow_wrap_interpreter_wrapper.so: undefined symbol: _ZN6tflite12tensor_utils24NeonVectorScalarMultiplyEPKaifPf

I've been using previous version 1.13.1 without any issues. I am getting the same when I build it using docker. Any ideas?

System info: Raspbian 9, python 3.5.3

lhelontra commented 5 years ago

Hi @Flipajs , I thinking that issue: https://github.com/tensorflow/tensorflow/issues/24323

In tf 1.13.1, i fixed this issue with an patch, i will investigating this again. Thanks!

muncok commented 5 years ago

Hi @lhelontra

I have the same issue at 1.14.0.

Does it also need some patches?

Thank you

lhelontra commented 5 years ago

I'm trying to apply the same patch that applies in tf 1.13.1. Loads tflite model but when allocate data, segmentation fault was happened.

lhelontra commented 5 years ago

I'm fixed this issue with an patch. This patch will'be available in next branch.

lhelontra commented 5 years ago

Tested with mobilenet from mobilenet and code bellow:

import tensorflow as tf
interpreter = tf.lite.Interpreter(model_path="mobilenet_v1_0.25_128_quant.tflite")
interpreter.allocate_tensors()
print(interpreter.get_input_details())
print(interpreter.get_output_details())
muncok commented 5 years ago

Tested with mobilenet from mobilenet and code bellow:

import tensorflow as tf
interpreter = tf.lite.Interpreter(model_path="mobilenet_v1_0.25_128_quant.tflite")
interpreter.allocate_tensors()
print(interpreter.get_input_details())
print(interpreter.get_output_details())

It works well at tf 1.14.0 Thanks @lhelontra

Flipajs commented 5 years ago

I confirm. Works great. Thanks!