search

ARM-software / armnn

Arm NN ML Software. The code here is a read-only mirror of https://review.mlplatform.org/admin/repos/ml/armnn
https://developer.arm.com/products/processors/machine-learning/arm-nn
MIT License
1.15k stars 307 forks source link

Supported models in onnx_parser #616

Closed saswat0 closed 2 years ago

saswat0 commented 2 years ago

I was trying to run inference on models released by onnx but most of them run into errors. I've attached some of the errors here

RuntimeError: Unsupported operation LRN for node ''  at function LoadGraph [armnn-dist/armnn/src/armnnOnnxParser/OnnxParser.cpp:886]

RuntimeError: Tensor numDimensions must be greater than 0 at function CheckValidNumDimensions [armnn-dist/armnn/src/armnn/Tensor.cpp:287]

RuntimeError: Error parsing Convolution node: . The 'group'=4 parameter should be 1 or be equal to the channel of the input shape=24 (in NCHW format).  at function ParseConv [armnn-dist/armnn/src/armnnOnnxParser/OnnxParser.cpp:1751]

RuntimeError: Unsupported operation LRN for node ''  at function LoadGraph [armnn-dist/armnn/src/armnnOnnxParser/OnnxParser.cpp:886]

Is there any comprehensive list of supported onnx models for compiling with pyarmnn's onnxparser?

MikeJKelly commented 2 years ago

Hi @saswat0

We do not have a comprehensive list of supported ONNX models. Our ONNX support is very limited as we're focusing on TF Lite support. You can see the ONNX operators we support and some details of the models we've used for testing at:

https://arm-software.github.io/armnn/latest/parsers.xhtml#S5_onnx_parser

Best regards, Mike

saswat0 commented 2 years ago

@MikeJKelly Understood. In that case, for the tflite models, are all the models listed in this above page the officially tested ones? I'm looking to test and benchmark image classifications models on aarch64 boards

MikeJKelly commented 2 years ago

@saswat0

Yes, those are the models we currently test with.

saswat0 commented 2 years ago

Got you