Error training tensor-flow object detection api using google cloud VM. [ImportError: No module named 'tensorflow.python.eager']

0
votes

When training tensor-flow object detection api using the typical steps in google cloud VM. After configuring all the dependence, when i try to run the train.py script the error [ImportError: No module named 'tensorflow.python.eager'] poped up. I already trained using the same steps in my local PC without any errors. I couldn't find any solution related to this error.

System Info: gcloud VM; TensorFlow-GPU 1.3.0; Python 3.5; CUDA 8.0 /cuDNN 6.0:

script running command: 

$ python3 train.py --logtostderr --train_dir=training/ --pipeline_config_path=ssd_mobilenet_v1_lap.config

Error:

Traceback (most recent call last): File "train.py", line 49, in from object_detection import trainer File "/usr/local/lib/python3.5/dist-packages/object_detection-0.1-py3.5.egg/object_detection/trainer.py", line 33 , in from deployment import model_deploy File "/home/ragulh28/project/models/research/slim/deployment/model_deploy.py", line 106, in from tensorflow.python.eager import context ImportError: No module named 'tensorflow.python.eager'

1
python-3.xtensorflowgcloud

1 Answers

0
votes

This issue is caused by a dependency on the new TF Eager API some of the newer models in slim use. They require the latest version of tensorflow, which is why the library is not being found.

Our apologies for the inconvenience. As a workaround, could you try checking out an older version of the Tensorflow Object Detection API? This commit should be a good candidate.