Inference on area of interest only

Hi All,

I am thinking for a solution to make inference only on the area of interest not on a full screen(need to perform inference only on the set area coordinates).
How can this be achieved in Holoscan ?

Hi there, sorry for the late reply, at this moment, you could add your custom operator to do the cropping of the frame (simplest would be in Numpy, Cupy or other libraries), and pass the cropped frame downstream to the inference operator.

1 Like

Thank you, Can this be applied for livestream videos as well? Please suggest other libraries other than Cupy or Numpy

Yes this can be applied for live stream frames too. You can choose any GPU accelerated libraries you are familiar with, in the C++ API or Python API, it doesn’t need to be Cupy or Numpy. Please see Creating Operators - NVIDIA Docs

This holoscan::Tensor class is a wrapper around the DLManagedTensorCtx struct holding a DLManagedTensor object. As such, it provides a primary interface to access Tensor data and is interoperable with other frameworks that support the DLPack interface.

This holoscan.core.Tensor class supports both DLPack and NumPy’s array interface (__array_interface__ and __cuda_array_interface__ ) so that it can be used with other Python libraries such as CuPy, PyTorch, JAX, TensorFlow, and Numba. See the interoperability section for more details.