Marvin Teichmann
Marvin Teichmann
If I understand this script correctly, it calls demo.py on a whole bunch of images. This is **extremely** slow as the model is build and compiled for each example. Instead,...
Thanks! I am rather busy right now and this is a commit I would like to review closer to ensure that it does not infer with existing experiments. I will...
Looks like a linking error. Have you tried running make again? Or use a different gcc / cython version.
Yes, if you use a large filter size you will get a model very similar to FullCRF. Note that there are some subtle differences. FullCRF uses the Permutohedral Lattice approximation,...
DenseCRF uses log on input data (i.e. this [happens here](https://github.com/lucasb-eyer/pydensecrf/blob/master/pydensecrf/utils.py#L33-L34) ). Using logsoftmax instead of first log and then softmax is numerical more stable. Using the log does not seam...
Strange. What is your validation score?
What is the output of your evaluation scripts?
Thanks. You do refer to this [line](https://github.com/MarvinTeichmann/ConvCRF/blob/master/convcrf/convcrf.py#L583), right? I will clean up the code. The condition should be removed (hence always be false) since I am already using logsoftmax before.
No, this seams to be a bug.
Implement the network and adapt this [line](https://github.com/MarvinTeichmann/KittiBox/blob/master/hypes/kittiBox.json#L4).