Yolox-nano tfl.strided_slice op not supported on DarwiNN

[nlbutts]

Description

I'm trying to run a YOLOX-Nano model on the Coral USB device. When I run the edgetpu_compiler I get the following error messages:

Edge TPU Compiler version 16.0.384591198
Started a compilation timeout timer of 180 seconds.
loc("onnx_tf_prefix_/backbone/backbone/stem/Slice_21"): error: 'tfl.strided_slice' op is not supported on DarwiNN. See the operation detail:
	%9 = "tfl.strided_slice"(%7, %0, %2, %1) {begin_mask = 0 : i32, ellipsis_mask = 0 : i32, end_mask = 0 : i32, new_axis_mask = 0 : i32, shrink_axis_mask = 0 : i32} : (tensor<1x3x160x320x!quant.uniform<i8:f32, 1.000000e+00:-128>>, tensor<4xi32>, tensor<4xi32>, tensor<4xi32>) -> tensor<1x3x160x160x!quant.uniform<i8:f32, 1.000000e+00:-128>>
loc("onnx_tf_prefix_/backbone/backbone/stem/Slice_12"): error: 'tfl.strided_slice' op is not supported on DarwiNN. See the operation detail:
	%8 = "tfl.strided_slice"(%7, %3, %2, %1) {begin_mask = 0 : i32, ellipsis_mask = 0 : i32, end_mask = 0 : i32, new_axis_mask = 0 : i32, shrink_axis_mask = 0 : i32} : (tensor<1x3x160x320x!quant.uniform<i8:f32, 1.000000e+00:-128>>, tensor<4xi32>, tensor<4xi32>, tensor<4xi32>) -> tensor<1x3x160x160x!quant.uniform<i8:f32, 1.000000e+00:-128>>
loc("onnx_tf_prefix_/backbone/backbone/stem/Slice_4"): error: 'tfl.strided_slice' op is not supported on DarwiNN. See the operation detail:
	%12 = "tfl.strided_slice"(%11, %3, %2, %1) {begin_mask = 0 : i32, ellipsis_mask = 0 : i32, end_mask = 0 : i32, new_axis_mask = 0 : i32, shrink_axis_mask = 0 : i32} : (tensor<1x3x160x320x!quant.uniform<i8:f32, 1.000000e+00:-128>>, tensor<4xi32>, tensor<4xi32>, tensor<4xi32>) -> tensor<1x3x160x160x!quant.uniform<i8:f32, 1.000000e+00:-128>>
loc("onnx_tf_prefix_/backbone/backbone/stem/Slice_5"): error: 'tfl.strided_slice' op is not supported on DarwiNN. See the operation detail:
	%13 = "tfl.strided_slice"(%11, %0, %2, %1) {begin_mask = 0 : i32, ellipsis_mask = 0 : i32, end_mask = 0 : i32, new_axis_mask = 0 : i32, shrink_axis_mask = 0 : i32} : (tensor<1x3x160x320x!quant.uniform<i8:f32, 1.000000e+00:-128>>, tensor<4xi32>, tensor<4xi32>, tensor<4xi32>) -> tensor<1x3x160x160x!quant.uniform<i8:f32, 1.000000e+00:-128>>
loc("main"): error: dwl-check-illegal-ops failed: found illegal ops in func main: tfl.strided_slice
ERROR: Restored original execution plan after delegate application failure.
Compilation failed: Internal error.
Compilation child process completed within timeout period.
Compilation failed! 

My question is did I convert the Pytorch model to tflite incorrectly or is the model architecture fundamentally not support on a Coral device.

Click to expand!

Issue Type

Support

Operating System

Ubuntu

Coral Device

USB Accelerator

Other Devices

No response

Programming Language

Python 3.8

Relevant Log Output

See above.