hkust-vgd / pointwise

Code for Pointwise Convolutional Neural Networks, CVPR 2018
http://pointwise.scenenn.net
MIT License
130 stars 26 forks source link

tensorflow TypeError when applying the GPU conv3p operator #5

Closed YukiRain closed 6 years ago

YukiRain commented 6 years ago

I'm trying to use the GPU version of the conv3p operator, so I set the parameter "use_gpu" to true in the file "param.json". However, the following issue occurred when running the Python code "train_modelnet40_acsd.py": /python3.5/site-packages/tensorflow/python/framework/ops.py TypeError: Using atf.Tensoras a Pythonboolis not allowed. Useif t is not None:instead ofif t:to test if a tensor is defined, and use TensorFlow ops such as tf.cond to execute subgraphs conditioned on the value of a tensor. It seems that this may not be a problem in the Python codes. Is there anything wrong with the GPU compilation, or I'm incorrectly use the codes?

songuke commented 6 years ago

When compiling the custom convolution operator, did you execute ./tf_conv3p_compile.sh -a with the -a flag to activate atrous convolution?

The signature of the non-atrous and atrous convolution are different.

By default, the network model in train_modelnet40_acsd assumes atrous convolution.

YukiRain commented 6 years ago

Thanks!!! My neglicence. Very interesting project and nice idea. I learn a lot from the CUDA codes. 👍