https://manaai.cn

然后切换到　solov2 分支．
就可以运行这行命令实现转换：

python3 demo/export_onnx.py \
    --config-file configs/SOLOv2/coco/R50_lite_3x.yaml \
    --video-input 0.mp4 \
    --opts MODEL.WEIGHTS ./output/model_final.pth

请注意，实现记得下载好你的final.pth．　其次记得将不要修改代码，因为正常下是没有问题的：

最后我们的环境是：

torch                         1.7.1                     

当然更高版本应该问题不大．

进一步的，你转好了onnx之后，还可以将模型转成trt的文件进行trt的推理．

jim@jim-AERO-15-X9:~/solov2_d2$ python3 demo/export.py --config-file configs/SOLOv2/R101_3x.yaml --video-input 0.mp4 --opts MODEL.WEIGHTS ./output/model_final.pth
[06/24 16:24:40 detectron2]: Arguments: Namespace(confidence_threshold=0.5, config_file='configs/SOLOv2/R101_3x.yaml', input=None, opts=['MODEL.WEIGHTS', './output/model_final.pth'], output=None, video_input='0.mp4', webcam=False)
[06/24 16:24:44 fvcore.common.checkpoint]: [Checkpointer] Loading from ./output/model_final.pth ...
[06/24 16:24:45 fvcore.common.checkpoint]: [Checkpointer] Loading from ./output/model_final.pth ...
/home/jim/.local/lib/python3.8/site-packages/torch/tensor.py:587: RuntimeWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).
warnings.warn('Iterating over a tensor might cause the trace to be incorrect. '
Traceback (most recent call last):
File "demo/export.py", line 153, in
torch.onnx.export(model, inp, 'solov2.onnx', output_names={
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/init.py", line 271, in export
return utils.export(model, args, f, export_params, verbose, training,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 88, in export
_export(model, args, f, export_params, verbose, training, input_names, output_names,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 694, in _export
_model_to_graph(model, args, verbose, input_names,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 457, in _model_to_graph
graph, params, torch_out, module = _create_jit_graph(model, args,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 420, in _create_jit_graph
graph, torch_out = _trace_and_get_graph_from_model(model, args)
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 380, in _trace_and_get_graph_from_model
torch.jit._get_trace_graph(model, args, strict=False, _force_outplace=False, _return_inputs_states=True)
File "/home/jim/.local/lib/python3.8/site-packages/torch/jit/_trace.py", line 1139, in _get_trace_graph
outs = ONNXTracedModule(f, strict, _force_outplace, return_inputs, _return_inputs_states)(*args, **kwargs)
File "/home/jim/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/jim/.local/lib/python3.8/site-packages/torch/jit/_trace.py", line 125, in forward
graph, out = torch._C._create_graph_by_tracing(
File "/home/jim/.local/lib/python3.8/site-packages/torch/jit/_trace.py", line 116, in wrapper
outs.append(self.inner(*trace_inputs))
File "/home/jim/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 887, in _call_impl
result = self._slow_forward(*input, **kwargs)
File "/home/jim/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 860, in _slow_forward
result = self.forward(*input, **kwargs)
File "/home/jim/AdelaiDet/adet/modeling/solov2/solov2.py", line 108, in forward
images = self.preprocess_image(batched_inputs)
File "/home/jim/AdelaiDet/adet/modeling/solov2/solov2.py", line 152, in preprocess_image
images = [x["image"].to(self.device) for x in batched_inputs]
File "/home/jim/AdelaiDet/adet/modeling/solov2/solov2.py", line 152, in
images = [x["image"].to(self.device) for x in batched_inputs]
IndexError: too many indices for tensor of dimension 3

Hi
在solov2_d2下面没有找到export_onnx.py，只有export.py，而且运行之后报了这样一个错：
jim@jim-AERO-15-X9:~/solov2_d2$ python3 demo/export.py --config-file configs/SOLOv2/R101_3x.yaml --video-input 0.mp4 --opts MODEL.WEIGHTS ./output/model_final.pth
[06/24 16:24:40 detectron2]: Arguments: Namespace(confidence_threshold=0.5, config_file='configs/SOLOv2/R101_3x.yaml', input=None, opts=['MODEL.WEIGHTS', './output/model_final.pth'], output=None, video_input='0.mp4', webcam=False)
[06/24 16:24:44 fvcore.common.checkpoint]: [Checkpointer] Loading from ./output/model_final.pth ...
[06/24 16:24:45 fvcore.common.checkpoint]: [Checkpointer] Loading from ./output/model_final.pth ...
/home/jim/.local/lib/python3.8/site-packages/torch/tensor.py:587: RuntimeWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).
warnings.warn('Iterating over a tensor might cause the trace to be incorrect. '
Traceback (most recent call last):
File "demo/export.py", line 153, in
torch.onnx.export(model, inp, 'solov2.onnx', output_names={
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/init.py", line 271, in export
return utils.export(model, args, f, export_params, verbose, training,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 88, in export
_export(model, args, f, export_params, verbose, training, input_names, output_names,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 694, in _export
_model_to_graph(model, args, verbose, input_names,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 457, in _model_to_graph
graph, params, torch_out, module = _create_jit_graph(model, args,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 420, in _create_jit_graph
graph, torch_out = _trace_and_get_graph_from_model(model, args)
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 380, in _trace_and_get_graph_from_model
torch.jit._get_trace_graph(model, args, strict=False, _force_outplace=False, _return_inputs_states=True)
File "/home/jim/.local/lib/python3.8/site-packages/torch/jit/_trace.py", line 1139, in _get_trace_graph
outs = ONNXTracedModule(f, strict, _force_outplace, return_inputs, _return_inputs_states)(*args, **kwargs)
File "/home/jim/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/jim/.local/lib/python3.8/site-packages/torch/jit/_trace.py", line 125, in forward
graph, out = torch._C._create_graph_by_tracing(
File "/home/jim/.local/lib/python3.8/site-packages/torch/jit/_trace.py", line 116, in wrapper
outs.append(self.inner(*trace_inputs))
File "/home/jim/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 887, in _call_impl
result = self._slow_forward(*input, **kwargs)
File "/home/jim/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 860, in _slow_forward
result = self.forward(*input, **kwargs)
File "/home/jim/AdelaiDet/adet/modeling/solov2/solov2.py", line 108, in forward
images = self.preprocess_image(batched_inputs)
File "/home/jim/AdelaiDet/adet/modeling/solov2/solov2.py", line 152, in preprocess_image
images = [x["image"].to(self.device) for x in batched_inputs]
File "/home/jim/AdelaiDet/adet/modeling/solov2/solov2.py", line 152, in
images = [x["image"].to(self.device) for x in batched_inputs]
IndexError: too many indices for tensor of dimension 3

@刘看山

请安装我们的adet．你如果系统有其他版本就卸载．我们的版本和其他版本不一样，再次声明，我们的代码都是经过修改过的版权代码，其他地方拷贝的代码报的错，不要与我们的教程混淆，请另外发帖．

/usr/bin/python3.8 /home/jim/solov2_d2/demo/export.py --config-file ../configs/SOLOv2/R101_3x.yaml --video-input ../0.mp4 --opts MODEL.WEIGHTS ../output/model_final.pth
[06/24 16:49:06 detectron2]: Arguments: Namespace(confidence_threshold=0.5, config_file='../configs/SOLOv2/R101_3x.yaml', input=None, opts=['MODEL.WEIGHTS', '../output/model_final.pth'], output=None, video_input='../0.mp4', webcam=False)
INFO 06.24 16:49:07 solov2.py:78: instance_shapes: [ShapeSpec(channels=256, height=None, width=None, stride=4), ShapeSpec(channels=256, height=None, width=None, stride=8), ShapeSpec(channels=256, height=None, width=None, stride=16), ShapeSpec(channels=256, height=None, width=None, stride=32), ShapeSpec(channels=256, height=None, width=None, stride=64)]
[06/24 16:49:15 fvcore.common.checkpoint]: [Checkpointer] Loading from ../output/model_final.pth ...
INFO 06.24 16:49:17 solov2.py:78: instance_shapes: [ShapeSpec(channels=256, height=None, width=None, stride=4), ShapeSpec(channels=256, height=None, width=None, stride=8), ShapeSpec(channels=256, height=None, width=None, stride=16), ShapeSpec(channels=256, height=None, width=None, stride=32), ShapeSpec(channels=256, height=None, width=None, stride=64)]
[06/24 16:49:18 fvcore.common.checkpoint]: [Checkpointer] Loading from ../output/model_final.pth ...
[WARN] exporting onnx...
batched_inputs: torch.Size([1, 3, 704, 736])
/home/jim/.local/lib/python3.8/site-packages/torch/nn/functional.py:3454: UserWarning: Default upsampling behavior when mode=bilinear is changed to align_corners=False since 0.4.0. Please specify align_corners=True if the old behavior is desired. See the documentation of nn.Upsample for details.
warnings.warn(
/home/jim/.local/lib/python3.8/site-packages/torch/nn/functional.py:3502: UserWarning: The default behavior for interpolate/upsample with float scale_factor changed in 1.6.0 to align with other frameworks/libraries, and now uses scale_factor directly, instead of relying on the computed output size. If you wish to restore the old behavior, please set recompute_scale_factor=True. See the documentation of nn.Upsample for details.
warnings.warn(
pred_masks /home/jim/.local/lib/python3.8/site-packages/torch/tensor.py:587: RuntimeWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).
warnings.warn('Iterating over a tensor might cause the trace to be incorrect. '
tensor([[[[0.0000e+00, 0.0000e+00, 0.0000e+00, ..., 3.9518e+00,
3.7653e+00, 3.8083e+00],
[0.0000e+00, 0.0000e+00, 0.0000e+00, ..., 4.5552e+00,
4.3284e+00, 4.1046e+00],
[0.0000e+00, 0.0000e+00, 0.0000e+00, ..., 4.7800e+00,
4.6571e+00, 4.3812e+00],
...,
[1.1723e-01, 1.5662e-01, 1.9572e-01, ..., 1.9446e-01,
1.1960e-01, 5.0942e-02],
[1.0176e-01, 1.4805e-01, 1.7128e-01, ..., 1.8052e-01,
1.1186e-01, 5.3612e-02],
[9.6826e-02, 9.2554e-02, 9.1506e-02, ..., 9.5873e-02,
6.4562e-02, 5.5044e-02]],

     [[0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      ...,
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 3.7478e-02,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 4.8736e-02,
       7.6189e-03, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 7.7488e-02,
       5.2618e-02, 2.4377e-02]],

     [[0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      ...,
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00]],

     ...,

     [[0.0000e+00, 0.0000e+00, 2.5016e-02,  ..., 2.8171e+01,
       2.9437e+01, 2.7834e+01],
      [0.0000e+00, 4.3195e-02, 8.7966e-02,  ..., 2.8479e+01,
       3.1145e+01, 3.0990e+01],
      [0.0000e+00, 5.1823e-02, 8.0432e-02,  ..., 2.8781e+01,
       3.2091e+01, 3.2862e+01],
      ...,
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 1.4385e-01,
       1.4950e-01, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 5.5557e-02,
       8.0213e-02, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00]],

     [[0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 2.4763e-02, 6.4760e-02,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 4.8006e-02, 4.6316e-02,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      ...,
      [8.3648e-03, 5.4830e-02, 1.3018e-01,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 7.6539e-03, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00]],

     [[4.6343e-02, 1.1665e-01, 1.7521e-01,  ..., 3.3670e+01,
       3.3471e+01, 2.9798e+01],
      [7.0594e-02, 1.5725e-01, 2.2316e-01,  ..., 3.2300e+01,
       3.3655e+01, 3.1853e+01],
      [8.3667e-02, 1.7335e-01, 2.1356e-01,  ..., 3.2309e+01,
       3.4434e+01, 3.3425e+01],
      ...,
      [0.0000e+00, 1.1504e-02, 3.4751e-02,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00],
      [0.0000e+00, 0.0000e+00, 0.0000e+00,  ..., 0.0000e+00,
       0.0000e+00, 0.0000e+00]]]], device='cuda:0') torch.Size([1, 256, 176, 184])

tensor(False, device='cuda:0')
tensor(False, device='cuda:0')
pred cate: torch.Size([3872, 80])
pred kernel: torch.Size([3872, 256])
/home/jim/solov2_d2/adet/modeling/solov2/solov2.py:628: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
assert not torch.isnan(seg_preds).any(), 'seg_preds contains nan'
/home/jim/solov2_d2/adet/modeling/solov2/solov2.py:656: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
sh = torch.tensor(seg_preds.shape)
/home/jim/solov2_d2/adet/modeling/solov2/solov2.py:658: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
sh_kernel = torch.tensor(kernel_preds.shape)
/home/jim/solov2_d2/adet/modeling/solov2/solov2.py:666: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
seg_masks = seg_preds > torch.tensor(self.mask_threshold).float()
/home/jim/solov2_d2/adet/modeling/solov2/solov2.py:689: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
keep = cate_scores >= torch.tensor(
/home/jim/solov2_d2/adet/modeling/solov2/solov2.py:745: TracerWarning: Converting a tensor to a Python integer might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
size=(max(int(ori_h0.6), 736), max(int(ori_w0.6), 992)),
/home/jim/solov2_d2/adet/modeling/solov2/solov2.py:747: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
seg_masks = seg_masks > torch.tensor(self.mask_threshold).float()
Traceback (most recent call last):
File "/home/jim/solov2_d2/demo/export.py", line 153, in
torch.onnx.export(model, inp, 'solov2.onnx', output_names={
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/init.py", line 271, in export
return utils.export(model, args, f, export_params, verbose, training,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 88, in export
_export(model, args, f, export_params, verbose, training, input_names, output_names,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 694, in _export
_model_to_graph(model, args, verbose, input_names,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 463, in _model_to_graph
graph = _optimize_graph(graph, operator_export_type,
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 206, in _optimize_graph
graph = torch._C._jit_pass_onnx(graph, operator_export_type)
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/init.py", line 309, in _run_symbolic_function
return utils._run_symbolic_function(*args, **kwargs)
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 993, in _run_symbolic_function
symbolic_fn = _find_symbolic_in_registry(domain, op_name, opset_version, operator_export_type)
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/utils.py", line 950, in _find_symbolic_in_registry
return sym_registry.get_registered_op(op_name, domain, opset_version)
File "/home/jim/.local/lib/python3.8/site-packages/torch/onnx/symbolic_registry.py", line 116, in get_registered_op
raise RuntimeError(msg)
RuntimeError: Exporting the operator linspace to ONNX opset version 11 is not supported. Please feel free to request support or submit a pull request on PyTorch GitHub.

Process finished with exit code 1