RedHair

Hi @Alexey-Kamenev thanks for your reply. Since there're no weights in ELU layer, I am wondering how to serialize weight. To be more specific, do all layer variables need serialization?...

Hi, your answer really saves my time. Do you mean I just need to serialize those important variables and recreate less important variables during deserialization?

I think the author just calculates the average of each matrix over all images. The calibration matrices for each image and lidar can be downloaded from [kitti website](http://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d).

if you want to detect objects in lidar, you don't need those matrices. calibration is needed because the label in kitti is in camera coordinate.

No, you can't train all the classes without modifying the code.

The parameters for car and pedestrian/cyclist are a bit different in the paper. However you can assume they are the same.

@traveller59 Did you preprocess the angle in other way? or you just disabled the angle limit function?

@traveller59 thanks for your swift reply. According to readme file in development kit, it seems the angle is set within [-pi, pi] by default. Do you do anything else or...

Hi, I use the newest version. I have written a Conv3DTranspose layer and uploaded. You can check it out in code directory.

Hi, I haven't reproduced the result because the the whole training set in SceneFlow is too large. I am still trying to figure out why the author needs only two...