Benjamin Ummenhofer
Benjamin Ummenhofer
I think consistency for the depth needs to be enforced for both mappings. This could be done with a special loss on the respective borders. In the beginning of this...
We did not experiment with depth prediction, just camera motion estimation and in that case a simple feed forward network was able to deal with the distortions. This does not...
Yes, it seems to be about Cython and numpy. Did you try to update numpy or Cython?
Can you check if the tensorflow include directory is properly set. This information is printed during the cmake configure step. Also check if you are using a recent cmake version,...
```TENSORFLOW_FRAMEWORK_LIB``` must point to the ```tensorflow_framework.so``` If you use a virtualenv make sure to activate it before invoking cmake. The instructions in the README.md worked for us on ubuntu 18.04...
It means that the training has to wait for data and could be faster.
Sure, please create a PR.
> BTW, it's better to change submodule lmbspecialops @ 06fd1a4 to the latest commit to catch up with the changes in flow_to_depth2. Thanks! I updated the submodule > What kind...
I think the problem could be in tfutils in the ```average_gradients``` or ```combine_loss_dicts``` function. Unfortunately, I don't have time right now to investigate. You are right, the l1 loss for...
The timestep is a hyperparameter and can be changed in the constructor of the model class. The training data should be compatible to the timestep usded for the advection step...