Benjamin Ummenhofer
Benjamin Ummenhofer
We have now added a ```flow_to_depth2``` op with bugfixes. The data in ```gt.zip``` however, still looks strange. Does the scale of the depth value correspond to the unit translation?
You simply scale the depth (not inverse depth) with the same scalar factor as the translation.
If you are using tensorflow 1.4 try updating demon and lmbspecialops to the latest commits.
Yes, this looks now like an ABI problem. Please have a look at the Issue mentioned by @JiamingSuen Try to start with a clean build directory and use a recent...
The dependencies for the lmbspecialops library have changed. The missing member was introduced after tensorflow 1.0.0. Building lmbspecialops should work with tensorflow 1.3 or later. I will update the documentation...
It is possible, but we did not test it. A problem is the tensor data format. Our custom ops that we use e.g. flow_to_depth2 and depth_to_flow only support the NCHW...
The sculptur_*txt files are the ground truth. They are used in the create_dataset_and_use_readerop.py example.
Code for training is in the *devel* branch.
Can you check if ```import lmbspecialops``` works in your python interpreter? Does it print the correct path to the .so file?
Did you compile the data reader? Does it print the path of the used .so file during import?