Memory Usage Question

[liamwebsterxyz]

Is my understanding correct that this library is applicable in the following scenario? Given a stream, this library utilizes the available GPU to decode the stream and provides a tensor located on the GPU, which references the decoded image frame. This tensor can then be used by another model on the GPU.

[BykadorovR]

Hello,
Yes, exactly, the whole processing (decoding, post-processing) happens on GPU and then the result is converted to Pytorch Tensor:

tensor-stream/src/Wrappers/WrapperPython.cpp

Line 322 in ba0df3c

outputTensor = torch::from_blob(processedFrame->opaque, { processedFrame->height, processedFrame->width, (int) channels },