Can the hailonet gstreamer plugin accept input tensors that have more than 3 channels?

General Discussion
,
1

Hi community.

I have a hef model that takes an input tensor with dimensions (10, 8, 4096) (H, W, C) In my gstreamer pipeline the upstream element sets the caps as (see gstreamer pipeline image below)

other/tensor, type=uint8, dims=(string)10:8:4096

image
image1032×372 46.7 KB

But I get the error

Features of input activ/layer_1 is not 3 for RGB format (features=4096)

Is there a way to tell the hailonet gstreamer element that input has more than 3 channels. (I did not see any options in this code

(I am able to successfully run the model using the python API in hailo_python)

Thanks for any help

2

Hey @D_Fidelt,

Good question! The GStreamer element is actually built pretty tightly around standard video inputs (mostly RGB), so that’s more of a design constraint than anything to do with the Python API itself.

For non-standard tensor shapes like what you’re working with, the Python API is definitely the way to go – and it sounds like you’ve already got that working, which is great!

I’ll check with our R&D team to see if we can add more flexibility to the GStreamer element in an upcoming release, but for now I’d recommend sticking with the Python API (or the C++ API if you need that level of control).

Let me know if you run into any other issues!