Failed to install Tensorflow object detection API on TX2

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

Hello everyone,

Recently, I followed the official instructios (https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md) to install Tensorflow object detection API on TX2, I experienced a failure when compiling the .proto files at the following step.

# From tensorflow/models/research/
protoc object_detection/protos/*.proto --python_out=.'

I suspect my protoc version is at fault, but has no proof. For those of you who were able to install Tensorflow object detection API on TX2, what version of protoc did you use?
Are there any special steps besided the official instructions?

Thanks in advance.

2

Hi chenyu36,

Please refer to below topics:
[url]https://devtalk.nvidia.com/default/topic/1038957/jetson-tx2/tensorflow-for-jetson-tx2-/[/url]

[url]https://devtalk.nvidia.com/default/topic/1037019/jetson-tx2/tensorflow-object-detection-and-image-classification-accelerated-for-nvidia-jetson/[/url]

Thanks

3

Hi,
I think I found the solution to my issue.
I want to clarify that I did have Tensorflow installed on TX2 successfully before posting this topic.

Having Tensorflow installed is not the same as having Tensorflow Object Detection API installed.
I believe the latter depends on the former being installed first.

The models I used are from release 1.11
https://github.com/tensorflow/models/releases/tag/v1.11

I have to download the latest protobuf which is 3.6.1 as of this writing, using the link
https://github.com/protocolbuffers/protobuf/releases/download/v3.6.1/protoc-3.6.1-linux-aarch_64.zip
Note: protoc-3.6.1-linux-aarch_64.zip is for linux-aarch_64 architecture, instead of the linux-x86.

Save the protoc-3.6.1-linux-aarch_64.zip to tensorflow/models/research/

# From tensorflow/models/research/
unzip protoc-3.6.1-linux-aarch_64.zip

Run the compilation process

# From tensorflow/models/research/
./bin/protoc object_detection/protos/*.proto --python_out=.

The remaining process is based on the official instructions:

# From tensorflow/models/research/
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim

Test that Tensorflow Object Detection API is correctly installed by running the following command:

# From tensorflow/models/research/
python3 object_detection/builders/model_builder_test.py

Here are my test results.

................./usr/local/lib/python3.5/dist-packages/tensorflow/python/util/tf_inspect.py:75: DeprecationWarning: inspect.getargspec() is deprecated, use inspect.signature() instead
  return _inspect.getargspec(target)
.....
----------------------------------------------------------------------
Ran 22 tests in 0.449s

OK