chunxt
commented
1 year ago I changed the 'b'(bactsize) to 4 in "toodhead.py", line 252, in forward anchor = torch.cat([anchor for in range(b)]) about the above issue. But another problem is occured. I could not find where set the bactsize in "channel_analyzer.py". So "template_output = model(**inputs)" is error, because "Forward failed in loss mode as The size of tensor a (1568) must match the size of tensor b (784) at non-singleton dimension 0".
Checklist
Describe the question you meet
I have retest the detection demo retinaet_prue successfully, But when I try to test the anther mmdetection model 'TOOD', it failed. The information is "'Proxy' object cannot be interpreted as an integer". I have followed this issue "[Bug]torch.fx.proxy.TraceError: Proxy object cannot be iterated. #554", but it does not work.
The detail is below. File "tools/train.py", line 121, in
main()
File "tools/train.py", line 114, in main
runner = Runner.from_cfg(cfg)
File "/home/ry/anaconda3/envs/ry/lib/python3.8/site-packages/mmengine/runner/runner.py", line 465, in from_cfg
runner = cls(
File "/home/ry/anaconda3/envs/ry/lib/python3.8/site-packages/mmengine/runner/runner.py", line 413, in init
self.model = self.build_model(model)
File "/home/ry/anaconda3/envs/ry/lib/python3.8/site-packages/mmengine/runner/runner.py", line 839, in build_model
model = MODELS.build(model)
File "/home/ry/anaconda3/envs/ry/lib/python3.8/site-packages/mmengine/registry/registry.py", line 570, in build
return self.build_func(cfg, args, kwargs, registry=self)
File "/home/ry/anaconda3/envs/ry/lib/python3.8/site-packages/mmengine/registry/build_functions.py", line 232, in build_model_from_cfg
return build_from_cfg(cfg, registry, default_args)
File "/home/ry/anaconda3/envs/ry/lib/python3.8/site-packages/mmengine/registry/build_functions.py", line 121, in build_from_cfg
obj = obj_cls(args) # type: ignore
File "/home/ry/DLtcx/exp_master/mmrazor/mmrazor/implementations/pruning/group_fisher/algorithm.py", line 57, in init
self.mutator.prepare_from_supernet(self.architecture)
File "/home/ry/DLtcx/exp_master/mmrazor/mmrazor/models/mutators/channel_mutator/channel_mutator.py", line 113, in prepare_from_supernet
units = self._prepare_from_tracer(supernet, self.parse_cfg)
File "/home/ry/DLtcx/exp_master/mmrazor/mmrazor/models/mutators/channel_mutator/channel_mutator.py", line 311, in _prepare_from_tracer
unit_configs = tracer.analyze(model)
File "/home/ry/DLtcx/exp_master/mmrazor/mmrazor/models/task_modules/tracer/channel_analyzer.py", line 107, in analyze
fx_graph = self._fx_trace(model)
File "/home/ry/DLtcx/exp_master/mmrazor/mmrazor/models/task_modules/tracer/channel_analyzer.py", line 136, in _fx_trace
return self.tracer.trace(model, concrete_args=args)
File "/home/ry/DLtcx/exp_master/mmrazor/mmrazor/models/task_modules/tracer/fx_tracer.py", line 310, in trace
'output', (self.create_arg(fn(args)), ), {},
File "/home/ry/anaconda3/envs/ry/lib/python3.8/site-packages/torch/fx/_symbolic_trace.py", line 652, in flatten_fn
tree_out = root_fn(*tree_args)
File "/home/ry/DLtcx/exp_master/mmdetection/mmdet/models/detectors/base.py", line 103, in forward
return self._forward(inputs, data_samples)
File "/home/ry/DLtcx/exp_master/mmdetection/mmdet/models/detectors/single_stage.py", line 133, in _forward
results = self.bbox_head.forward(x)
File "/home/ry/DLtcx/exp_master/mmdetection/mmdet/models/dense_heads/toodhead.py", line 252, in forward
anchor = torch.cat([anchor for in range(b)])
TypeError: 'Proxy' object cannot be interpreted as an integer.