mbcel

Loss is NaN when using half precision

3

When I run my model on half precision(fp16) the Loss function returns NaN. It all works fine when I use normal floating point precision (fp32) so I don't think it...

Error when trying to use CudaHalfTensor for training

4

I want to train my model using fp16 precision on my gpu. My gpu has the Pascal architecture and the cutorch.hasHalf flag indicates true. I am using cuDNN 5.1 and...

Efficient way to create python list from JavaList or JavaArray

1

I want to pass a big float array (~ 30 000 floats) that is created on the java side to the python side. This all works quite efficient within about...

No callback on software trigger in C++

2

I am using the tcamcamera.cpp class from the examples to get an image on software trigger. For this I am initializing the camera as follows: ``` std::shared_ptr exposureProperty = tcamCamera.get_property("Exposure...

tiscamera does block std::cout prints

5

I am experiencing a strange behaviour when using the tiscamera library. When I start the camera via `TcamCamera::start() ` and after setting exposure properties somehow no `std::cout` prints are shown...

I am a little bit confused how the weighting is done of each input element of the deformed kernel in your implementation. In your implementation it looks like you are...

Understanding normalization of advantage function

7

I am wondering about the normalization of the advantage function in PPO. Before training on a batch the mean of the advantage function is subtracted and it's divided by its...

General understanding of why this works better than general optimization

2

Really interesting work! There is one question that came up to me about this general approach of modelling a scene with an MLP compared to a general normal optimization without...

I am wondering where you found the specification values regarding the imu of the Rt3003 in your paper. Here you say that the imu has a gyro and bias stability...

Difference of x, y coordinates of dual return mode

1

I am currently investigating the Dual Return mode for the Horizon Lidar and found that in the SDK the data field for such point (LivoxDualExtendRawPoint) does contain 2 x and...