https://community.nxp.com/t5/i-MX-Processors/IMX8M-nn-imx-NNAPI-1-3-support/m-p/1209232#M167495 <P>Currently the nn-imx (1.1.9) only support nnapi 1.2 version (Android api 29).</P><P>When will nn-imx support nnapi 1.3 (android api 30) ?<BR /><BR />Since nnapi1.2 have so many unsupported implementation.<BR />Many tensorflow lite neutral network model are fall back to CPU.</P><P>Noticeably Resize(with align_corners enabled), ANEURALNETWORKS_TENSOR_QUANT8_ASYMM_SIGNED etc.</P><P>Which makes the NPU running so slow, since many layers are fallback to CPU.</P> Sat, 09 Jan 2021 19:57:25 GMT keithmok 2021-01-09T19:57:25Z https://community.nxp.com/t5/i-MX-Processors/IMX8M-nn-imx-NNAPI-1-3-support/m-p/1209232#M167495 <P>Currently the nn-imx (1.1.9) only support nnapi 1.2 version (Android api 29).</P><P>When will nn-imx support nnapi 1.3 (android api 30) ?<BR /><BR />Since nnapi1.2 have so many unsupported implementation.<BR />Many tensorflow lite neutral network model are fall back to CPU.</P><P>Noticeably Resize(with align_corners enabled), ANEURALNETWORKS_TENSOR_QUANT8_ASYMM_SIGNED etc.</P><P>Which makes the NPU running so slow, since many layers are fallback to CPU.</P> Sat, 09 Jan 2021 19:57:25 GMT https://community.nxp.com/t5/i-MX-Processors/IMX8M-nn-imx-NNAPI-1-3-support/m-p/1209232#M167495 keithmok 2021-01-09T19:57:25Z