TFLite GPU Delegate error on Android

TFLite Movenet causing issue with gpu delegate on android.
It’s running fine with CPU, but ofcourse that’s very slow.
Error: (Been stuck for long, any help would be most appreciated :smiley: )

TFLite Warning: Following operations are not supported by GPU delegate:
ARG_MAX: Operation is not supported.
CAST: Not supported Cast case. Input type: UINT8 and output type: FLOAT32
CONCATENATION: OP is supported, but tensor type/shape isn’t compatible.
DEQUANTIZE:
FLOOR_DIV: OP is supported, but tensor type/shape isn’t compatible.
GATHER_ND: Operation is not supported.
MUL: OP is supported, but tensor type/shape isn’t compatible.
PACK: OP is supported, but tensor type/shape isn’t compatible.
RESHAPE: OP is supported, but tensor type/shape isn’t compatible.
SUB: OP is supported, but tensor type/shape isn’t compatible.
97 operations will run on the GPU, and the remaining 200 operations will run on the CPU.
UnityEngine.Debug:LogWarning (object)
TensorFlowLite.ErrorReporter:OnErrorReporter (intptr,string,intptr) (at ./Library/PackageCache/com.github.asus4.tflite@2.10.0/Runtime/ErrorReporter.cs:49)
TensorFlowLite.Interpreter:.ctor (byte,TensorFlowLite.InterpreterOptions)
TensorFlowLite.BaseImagePredictor1<sbyte>:.ctor (string,TensorFlowLite.BaseImagePredictor1/Accelerator)
TensorFlowLite.MoveNet.MoveNetSinglePose:.ctor (TensorFlowLite.MoveNet.MoveNetSinglePose/Options) (at Assets/MoveNet/MoveNetSinglePose.cs:26)
MoveNetSinglePoseSample:Start () (at Assets/MoveNet/MoveNetSinglePoseSample.cs:37)

1 Like

Hi, Could you please refer the comment here TFlite GPU delegate crash on Movenet · Issue #56795 · tensorflow/tensorflow · GitHub regarding the Movenet update on GPUdelegate which states as below.

"Movenet didn’t support GPU on Android and it’s non-trivial to do it. Unfortunately I don’t see it going to be supported anytime soon.

The good news is that we’re will have a different model architecture supporting GPU in coming months, so stay tuned."