How to convert detectron2 models into fx graph?

[seulkiyeom]

Hi. I'd like to convert the models from detectron2 into fx graph toward the implementation on edge device.

Is there any way to do it?

I slightly modified a code tools/deploy/export_model.py for exporting the detectron2 model into fx graph

including
from torch.fx import symbolic_trace
traced = symbolic_trace(traceable_model)

However, there is an error saying
Traceback (most recent call last):
File "/mnt/ssd2/skyeom/anaconda3/envs/py39/lib/python3.9/runpy.py", line 197, in _run_module_as_main
return _run_code(code, main_globals, None,
File "/mnt/ssd2/skyeom/anaconda3/envs/py39/lib/python3.9/runpy.py", line 87, in _run_code
exec(code, run_globals)
File "/mnt/ssd2/skyeom/.vscode-server/extensions/ms-python.python-
2022.16.1/pythonFiles/lib/python/debugpy/adapter/../../debugpy/launcher/../../debugpy/__main__.py", line 39, in <module>
cli.main()
File "/mnt/ssd2/skyeom/.vscode-server/extensions/ms-python.python-
2022.16.1/pythonFiles/lib/python/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py", line 430, in main
run()
File "/mnt/ssd2/skyeom/.vscode-server/extensions/ms-python.python-
2022.16.1/pythonFiles/lib/python/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py", line 284, in run_file
runpy.run_path(target, run_name="__main__")
File "/mnt/ssd2/skyeom/.vscode-server/extensions/ms-python.python-
2022.16.1/pythonFiles/lib/python/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 321, in run_path
return _run_module_code(code, init_globals, run_name,
File "/mnt/ssd2/skyeom/.vscode-server/extensions/ms-python.python-
2022.16.1/pythonFiles/lib/python/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 135, in _run_module_code
_run_code(code, mod_globals, init_globals,
File "/mnt/ssd2/skyeom/.vscode-server/extensions/ms-python.python-
2022.16.1/pythonFiles/lib/python/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 124, in _run_code
exec(code, run_globals)
File "/mnt/ssd2/skyeom/PycharmProjects/detectron2/tools/deploy/export_model.py", line 237, in <module>
exported_model = export_tracing(torch_model, sample_inputs)
File "/mnt/ssd2/skyeom/PycharmProjects/detectron2/tools/deploy/export_model.py", line 128, in export_tracing
traced = symbolic_trace(traceable_model)
File "/mnt/ssd2/skyeom/anaconda3/envs/py39/lib/python3.9/site-packages/torch/fx/_symbolic_trace.py", line 878, in symbolic_trace
graph = tracer.trace(root, concrete_args)
File "/mnt/ssd2/skyeom/anaconda3/envs/py39/lib/python3.9/site-packages/torch/fx/_symbolic_trace.py", line 587, in trace
self.create_node('output', 'output', (self.create_arg(fn(*args)),), {},
File "/mnt/ssd2/skyeom/PycharmProjects/detectron2/detectron2/export/flatten.py", line 282, in forward
inputs_orig_format = self.inputs_schema(args)
File "/mnt/ssd2/skyeom/PycharmProjects/detectron2/detectron2/export/flatten.py", line 91, in __call__
return tuple(super().__call__(values))
File "/mnt/ssd2/skyeom/PycharmProjects/detectron2/detectron2/export/flatten.py", line 73, in __call__
values = self._split(values, self.sizes)
File "/mnt/ssd2/skyeom/PycharmProjects/detectron2/detectron2/export/flatten.py", line 58, in _split
len(values) == expected_len
File "/mnt/ssd2/skyeom/anaconda3/envs/py39/lib/python3.9/site-packages/torch/fx/proxy.py", line 291, in __len__
raise RuntimeError("'len' is not supported in symbolic tracing by default. If you want "
RuntimeError: 'len' is not supported in symbolic tracing by default. If you want this call to be recorded, please call torch.fx.wrap('len') at module scope

Is there anybody who can figure out the problem? or know an alternative way to convert into fx graph?