I am trying to Implement a Instance segmentation custom model that can be run on Intel NCS2 for inference.
Upsampling layer is one crucial layer in generating masks, but not is not supported.
I need to know if you can recommend me to go with deconvolution layer and obtain similar functionality.
Also is there any chance of the same upsampling layer being supported in the future versions of the OpenVINO toolkit?
Thanks for reaching out. As this website says:
"Upsampling refers to any technique that, well, upsamples your image to a higher resolution".
It addresses some possible substitutions, such as Deconvolution and Resample layers. These two layers are supported by the VPU, as stated in the supported layers documentation.
About future releases, we cannot share information about it. Please check the Release notes for updates on each release.