Error while trying to run NiftyNet quick start command

Question

I'm trying out NiftyNet and got stuck at the first step. Trying to run the quickstart command

python net_download.py dense_vnet_abdominal_ct_model_zoo python net_segment.py inference -c ~/niftynet/extensions/dense_vnet_abdominal_ct/config.ini

gives me

KeyError: "Registering two gradient with name 'FloorMod' !(Previous registration was in _find_and_load_unlocked :955)"

Could any one help? I'm using Ubuntu 16.04 with Nvidia GPU. Tried tensorflow:1.4.1-py3 docker image, Anaconda with CPU version of tensorflow and native python with CPU version of tensorflow and I get the same error.

I'm pretty sure it's something I did wrong because I get the same error from those different environment but I'm not sure what...

Thanks!

Answer 1

I think this is a change from TensorFlow 1.3 (which is the official NiftyNet dependency) to 1.4. You can switch back to TF1.3, or comment out the following lines in niftynet/layer/resampler.py

@tf.RegisterGradient('FloorMod')
def _floormod_grad(op, grad):
    return [None, None]