Issues installing DIGITS on TX2

When installing the python requirements for the NVIDIA Caffe package there is an error on the leveldb package:

sudo pip install -r python/requirements.txt

In file included from ./leveldb/port/port_posix.h:50:0,
                     from ./leveldb/port/port.h:14,
                     from ./leveldb/db/filename.h:14,
                     from leveldb/db/builder.cc:7:
    ./leveldb/port/atomic_pointer.h:212:2: error: #error Please implement AtomicPointer for this platform.
     #error Please implement AtomicPointer for this platform.
      ^
    error: command 'aarch64-linux-gnu-gcc' failed with exit status 1

Anyone else run into this?

I was following the install instructions for DIGIT at GitHub - dusty-nv/jetson-inference: Hello AI World guide to deploying deep-learning inference networks and deep vision primitives with TensorRT and NVIDIA Jetson.

DIGITS isn’t intended to be installed on Jetson or officially supported on ARM. The setup directions indicate to install DIGITS on the host (x86_64). Please follow the System Setup directions from the tutorial to configure DIGITS on the host (i.e. you will also need to install CUDA Toolkit, NVIDIA driver, cuDNN, and NVcaffe for DIGITS). See this diagram illustrating the DIGITS workflow (i.e. train with DIGITS and deploying inference on Jetson):

I also met this problem (“pip install leveldb” failed) when I tried to install Caffe on Jetson TX2. Here is how I resolved it.

(Replace ‘python3’ with ‘python’, if you are using python 2.7.)

### Build and install leveldb-0.20
wget https://pypi.python.org/packages/03/98/1521e7274cfbcc678e9640e242a62cbcd18743f9c5761179da165c940eac/leveldb-0.20.tar.gz
tar xzvf leveldb-0.20.tar.gz
cd leveldb-0.20
python3 setup.py build
sudo python3 setup.py install

You can also refer to my blog post for a more detailed description about how I installed Caffe on Jetson TX2:

https://jkjung-avt.github.io/caffe-on-tx2/.

@jkjung13 I’m getting a bus error and seg fault after I compile and run DIGITS on the TX2. Did you compile DIGITS from source on your TX2?

I installed Caffe on Jetson TX2 mainly for inference only. I did not install DIGITS on Jetson TX2.

On the other hand, I do use NVIDIA DIGITS on my x86_64 PC: [url]https://jkjung-avt.github.io/deep-learning-pc/[/url]. I have a GeForce GTX-1080 in that PC, and it works pretty well. For example I have trained a DetectNet fish detector with it: [url]https://jkjung-avt.github.io/detectnet-training/[/url].

As dusty_nv has already replied, NVIDIA DIGITS does not officially support the Jetson TX2 platform. So you’d be on your own if you’d like to get that to work.