jgorgenucsd / corr_tf

An implementation of the FlowNetC correlation layer in tensorflow

corr_tf

An implementation of the FlowNetC correlation layer in tensorflow

The FlowNetC architecture (https://arxiv.org/abs/1504.06852) uses a novel cross correlation layer This is an implementation of that cross correlation layer in tensorflow, with CUDA support.

The function correlation_layer.corr expects two arguments, 4 dim tensors of size (batch_size,height,width,num_channels)

REQUIRES: Tensorflow >= 1.1, CUDA >= 8.0, CMAKE >= 2.8

BUILDING:

$ mkdir build

$ cd build

$ cmake ..

$ make

TESTING:

$ python correlation_tests.py

corr_tf's Issues

About video memory of get_correlation_layer in flownet.py

Hello, here I met such issues: I use your implementation in my own code ,my displacement value set 20, and stride set 1,I will get OOM error, if I decrease the two values, the OOM error disappears. May I ask if there is another implementation to avoid the video memory overflow issues.

Build error

Please check this why is giving error.

1D Correlation for Stereo Disparity Estimation

Hi, I am quite new to TensorFlow. I was wondering, why do we need CUDA implementation for this, I mean can't we code a correlation function using Python / TensorFlow APIs?

Also, for performing 1D correlation operation in case of stereo disparity estimation, is it okay to just compute dot products of corresponding elements of left / right feature maps (kernel size is thus, 1x1)?

./build/libcorrelation.so: undefined symbol: _ZTIN10tensorflow8OpKernelE

After "cmake .." and "make", when I run "python correlation_tests.py ", it reports error:

Traceback (most recent call last):
File "correlation_tests.py", line 11, in
import correlation_layer as cl
File "/home/yujiao/code/crossview_localisation/crop/corr_tf/correlation_layer.py", line 2, in
correlation_module = tf.load_op_library("./build/libcorrelation.so")
File "/home/yujiao/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/load_library.py", line 56, in load_op_library
lib_handle = py_tf.TF_LoadLibrary(library_filename)
tensorflow.python.framework.errors_impl.NotFoundError: ./build/libcorrelation.so: undefined symbol: _ZTIN10tensorflow8OpKernelE

May I ask what I can do in this case?