FAIL : This is an invalid model. Error: the graph is not acyclic.

[ywpkwon]

I have a ONNX model and I'm trying to compile using this tool. This model is not acyclic but the message says so.

Number of OD backbone nodes = 0 
Size of odBackboneNodeIds = 0 

Preliminary subgraphs created = 2 
Final number of subgraphs created are : 1, - Offloaded Nodes - 306, Total Nodes - 313 
Traceback (most recent call last):
  File "onnxrt_compiler_tsr.py", line 201, in <module>
    run_model(config)
  File "onnxrt_compiler_tsr.py", line 137, in run_model
    sess_options=so)
  File "/home/paul/.virtualenvs/tidl81/lib/python3.6/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 283, in __init__
    self._create_inference_session(providers, provider_options)
  File "/home/paul/.virtualenvs/tidl81/lib/python3.6/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 315, in _create_inference_session
    sess.initialize_session(providers, provider_options)
onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : This is an invalid model. Error: the graph is not acyclic.

The model itself should be okay since

1. There is no such error when I just use ['CPUExecutionProvider'].

2. Here is the tempDir/runtimes_visualization.svg. The backbone is based on mobileNet. There are 14 output tensors. The network is not acyclic.

3. When I compile manually with ti-processor-sdk-rtos-j721e-evm-08_01_00_13/tidl_j7_08_01_00_05/ti_dl/utils/tidlModelImport/out/tidl_model_import, I was able to compile the model into TIDL bin files.

Is this tool only for small example networks? The manual way (i.e, ./out/tidl_model_import.out ./custom/tidl_import_model.txt) is too cumbersome to setup, so I was going to use this tool.

Here is the onnx model. (I don't need any post-processing.) debug-b1.zip

I gave these options:

framework: 'MMDetection'
meta_arch_type: 3
meta_layers_names_list: ''
model_path: ''
model_type: 'od'
num_images: 3
od_type: 'SSD'
mean: [0, 0, 0]
std: [1., 1., 1.]
image_size: [512, 256]