Details:

• With this PR user is able to pass 3D tensor in HWC layout (single RGB image) to 4D model (NCHW). E.g.
  • 'ppp.tensor().set_layout("HWC"); ppp.model().set_layout("NCHW"); ppp.build()' This will add 'unsqueeze' operation to add 'N' dimension at beginning, and then adds 'transpose' from NHWC to NCHW

Limitations:

• User's tensor rank shall be 'smaller' than model's input. I.e. user is not able to convert NCDHW to NCHW even if 'D' dimension is '1'
• Currently supported for 'preprocessing' only. Post-processor doesn't support convert NCHW->HWC as not needed (user can always convert it to NHWC and data will be the same)

Test:

• Unit tests
• template tests with hard-coded references
• CPU + GPU reference test

Tickets:

• 75287