Why the input data shape is [4, 8, 3,192, 640] when the batch szie is 16?

[brummi]

Why the input data shape is [4, 8, 3,192, 640] when the batch szie is 16?

[rockywind]

Thank you very much!
Yes I am running in a system with 4 GPUs

[Brummi]

Hi,
sure.
Could you please specify where you observe these data shapes?

[rockywind]

Hi,
This is the code reference.

[Brummi]

Assuming these are the shapes of the images that the network receives from the data loader.
[n, v, c, h, w]
n = Batch size
v = Number of frames / camera views per sample (e.g. for KITTI-360 it's 2 timesteps x 2 cameras (forward, fisheye) x 2 stereo (left, right) = 8)
c = Number of color channels
h = Height
w = Width

The batch size is directly fed to the dataloader. So it might be that you are running a system with 4 GPUs and the code automatically distributes the batch onto those 4 GPUs.

[ruili3]

Thank you very much! Yes I am running in a system with 4 GPUs

Hi may I ask how you make code work on multiple GPUs? Did you follow the advice raised from here?

Thank you a lot for the answer!