To use Coral Edge TPU's Python API, you must use their bundled Interpreter and Delegate base classes as well. I think these SWIG wrappers are probably built and linked against a specific TensorFlow commit.
Using libedgetpu.so with tensorflow.lite.Interpreter, results in an error that's almost certainly raised by an incorrect SWIG typemap.
Unsupported data type in custom op handler: 0Node number 2 (EdgeTpuDelegateForCustomOp)
If edgetpu contributors are accepting patches, continue troubleshooting by grepping through a diff of all typemap revisions between the commit libedgetpu.so is linked against and the current 2.2.0 TensorFlow release.
To use Coral Edge TPU's Python API, you must use their bundled
Interpreter
andDelegate
base classes as well. I think these SWIG wrappers are probably built and linked against a specific TensorFlow commit.Using libedgetpu.so with tensorflow.lite.Interpreter, results in an error that's almost certainly raised by an incorrect SWIG typemap.
Unsupported data type in custom op handler: 0Node number 2 (EdgeTpuDelegateForCustomOp)
If edgetpu contributors are accepting patches, continue troubleshooting by grepping through a diff of all typemap revisions between the commit libedgetpu.so is linked against and the current 2.2.0 TensorFlow release.
closes #13 link https://github.com/google-coral/edgetpu/issues/124