@刘看山 请问您好使的时候用的是哪个版本的tensorrt呀？

@刘看山 您可以把你使用的pytorch, cuda,cudnn,tensorrt版本告诉我吗？我再试试

@刘看山 在 solov2生成好的onnx模型转tensorrt的时候报错，麻烦看一下谢谢！ 中说：

是一下 onnx-tensorrt 这个工具。

你好，我用了你说的这个工具，报错是差不多的：
onnx2trt solov2.onnx -o solov2.trt
Input filename: solov2.onnx
ONNX IR version: 0.0.7
Opset version: 11
Producer name: pytorch
Producer version: 1.10
Domain:
Model version: 0
Doc string:
Parsing model
[2022-05-07 07:05:49 WARNING] onnx2trt_utils.cpp:364: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
While parsing node number 283 [Range -> "966"]:
ERROR: builtin_op_importers.cpp:3350 In function importRange:
[8] Assertion failed: inputs.at(0).isInt32() && "For range operator with dynamic inputs, this version of TensorRT only supports INT32!"

你好，我用了你说的这个工具，报错是差不多的：
onnx2trt solov2.onnx -o solov2.trt

Input filename: solov2.onnx
ONNX IR version: 0.0.7
Opset version: 11
Producer name: pytorch
Producer version: 1.10
Domain:
Model version: 0
Doc string:

Parsing model
[2022-05-07 07:05:49 WARNING] onnx2trt_utils.cpp:364: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
While parsing node number 283 [Range -> "966"]:
ERROR: builtin_op_importers.cpp:3350 In function importRange:
[8] Assertion failed: inputs.at(0).isInt32() && "For range operator with dynamic inputs, this version of TensorRT only supports INT32!"

使用tensorrt 8.2 cuda 11.1 pytorch 1.10.1 opencv 4.5.5

我使用这个命令：
trtexec --onnx=solov2.onnx --saveEngine=solov2.trt
报错部分：
[05/06/2022-15:38:59] [E] [TRT] ModelImporter.cpp:776: --- End node ---
[05/06/2022-15:38:59] [E] [TRT] ModelImporter.cpp:779: ERROR: builtin_op_importers.cpp:3352 In function importRange:
[8] Assertion failed: inputs.at(0).isInt32() && "For range operator with dynamic inputs, this version of TensorRT only supports INT32!"
[05/06/2022-15:38:59] [E] Failed to parse onnx file
[05/06/2022-15:38:59] [I] Finish parsing network model
[05/06/2022-15:38:59] [E] Parsing model failed
[05/06/2022-15:38:59] [E] Failed to create engine from model.
[05/06/2022-15:38:59] [E] Engine set up failed
&&&& FAILED TensorRT.trtexec [TensorRT v8204] # trtexec --onnx=solov2.onnx --saveEngine=solov2.trt

您好，我使用这个命令：
python3 demo/export.py --config-file configs/SOLOv2/wood_dataset/R101_3x.yaml --video-input 0.mp4 --opts MODEL.WEIGHTS tools/output/model_final.pth

回复: SOLOv2转ONNX教程
[05/05 17:31:02 detectron2]: Arguments: Namespace(confidence_threshold=0.5, config_file='configs/SOLOv2/wood_dataset/R101_3x.yaml', input=None, opts=['MODEL.WEIGHTS', 'tools/output/model_final.pth'], output=None, video_input='0.mp4', webcam=False)
[05/05 17:31:05 fvcore.common.checkpoint]: [Checkpointer] Loading from tools/output/model_final.pth ...
[05/05 17:31:06 fvcore.common.checkpoint]: [Checkpointer] Loading from tools/output/model_final.pth ...
/home/xss/Software/AdelaiDet/adet/modeling/solov2/solov2.py:152: TracerWarning: 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).
images = [x["image"].to(self.device) for x in batched_inputs]
Traceback (most recent call last):
File "demo/export.py", line 153, in
torch.onnx.export(model, inp, 'solov2.onnx', output_names={
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/init.py", line 275, in export
return utils.export(model, args, f, export_params, verbose, training,
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 88, in export
_export(model, args, f, export_params, verbose, training, input_names, output_names,
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 689, in _export
_model_to_graph(model, args, verbose, input_names,
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 458, in _model_to_graph
graph, params, torch_out, module = _create_jit_graph(model, args,
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 422, in _create_jit_graph
graph, torch_out = _trace_and_get_graph_from_model(model, args)
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 373, in _trace_and_get_graph_from_model
torch.jit._get_trace_graph(model, args, strict=False, _force_outplace=False, _return_inputs_states=True)
File "/usr/local/lib/python3.8/dist-packages/torch/jit/_trace.py", line 1160, in _get_trace_graph
outs = ONNXTracedModule(f, strict, _force_outplace, return_inputs, _return_inputs_states)(*args, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/torch/jit/_trace.py", line 127, in forward
graph, out = torch._C._create_graph_by_tracing(
File "/usr/local/lib/python3.8/dist-packages/torch/jit/_trace.py", line 118, in wrapper
outs.append(self.inner(*trace_inputs))
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1039, in _slow_forward
result = self.forward(*input, **kwargs)
File "/home/xss/Software/AdelaiDet/adet/modeling/solov2/solov2.py", line 108, in forward
images = self.preprocess_image(batched_inputs)
File "/home/xss/Software/AdelaiDet/adet/modeling/solov2/solov2.py", line 152, in preprocess_image
images = [x["image"].to(self.device) for x in batched_inputs]
File "/home/xss/Software/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

我看了markdown，尝试这这么做：
python onnx/export_model_to_onnx.py --config-file configs/SOLOv2/wood_dataset/R101_3x.yaml --output R101_3x.onnx --opts MODEL.WEIGHTS tools/output/model_final.pth
但是报错了：
[05/05 17:09:44 detectron2]: load Model:
tools/output/model_final.pth
/home/xss/Software/AdelaiDet/adet/modeling/solov2/solov2.py:152: TracerWarning: 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).
images = [x["image"].to(self.device) for x in batched_inputs]
Traceback (most recent call last):
File "onnx/export_model_to_onnx.py", line 226, in
main()
File "onnx/export_model_to_onnx.py", line 212, in main
torch.onnx.export(
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/init.py", line 275, in export
return utils.export(model, args, f, export_params, verbose, training,
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 88, in export
_export(model, args, f, export_params, verbose, training, input_names, output_names,
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 689, in _export
_model_to_graph(model, args, verbose, input_names,
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 458, in _model_to_graph
graph, params, torch_out, module = _create_jit_graph(model, args,
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 422, in _create_jit_graph
graph, torch_out = _trace_and_get_graph_from_model(model, args)
File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 373, in _trace_and_get_graph_from_model
torch.jit._get_trace_graph(model, args, strict=False, _force_outplace=False, _return_inputs_states=True)
File "/usr/local/lib/python3.8/dist-packages/torch/jit/_trace.py", line 1160, in _get_trace_graph
outs = ONNXTracedModule(f, strict, _force_outplace, return_inputs, _return_inputs_states)(*args, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/torch/jit/_trace.py", line 127, in forward
graph, out = torch._C._create_graph_by_tracing(
File "/usr/local/lib/python3.8/dist-packages/torch/jit/_trace.py", line 118, in wrapper
outs.append(self.inner(*trace_inputs))
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1039, in _slow_forward
result = self.forward(*input, **kwargs)
File "/home/xss/Software/AdelaiDet/adet/modeling/solov2/solov2.py", line 108, in forward
images = self.preprocess_image(batched_inputs)
File "/home/xss/Software/AdelaiDet/adet/modeling/solov2/solov2.py", line 152, in preprocess_image
images = [x["image"].to(self.device) for x in batched_inputs]
File "/home/xss/Software/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
您好，我安装了cuda 11.3 pytorch 1.10，训练solov2_d2的时候报这个错误：

[04/29 13:43:52 d2.data.common]: Serialized dataset takes 0.71 MiB
WARNING [04/29 13:43:52 d2.solver.build]: SOLVER.STEPS contains values larger than SOLVER.MAX_ITER. These values will be ignored.
/usr/local/lib/python3.8/dist-packages/numpy/core/getlimits.py:499: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
setattr(self, word, getattr(machar, word).flat[0])
/usr/local/lib/python3.8/dist-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
return self._float_to_str(self.smallest_subnormal)
/usr/local/lib/python3.8/dist-packages/numpy/core/getlimits.py:499: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
setattr(self, word, getattr(machar, word).flat[0])
/usr/local/lib/python3.8/dist-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
return self._float_to_str(self.smallest_subnormal)
[04/29 13:43:54 fvcore.common.checkpoint]: [Checkpointer] Loading from detectron2://ImageNetPretrained/MSRA/R-101.pkl ...
[04/29 13:43:54 d2.checkpoint.c2_model_loading]: Renaming Caffe2 weights ......
[04/29 13:43:54 d2.checkpoint.c2_model_loading]: Following weights matched with submodule backbone.bottom_up:

WARNING [04/29 13:43:54 fvcore.common.checkpoint]: Some model parameters or buffers are not found in the checkpoint:
backbone.fpn_lateral2.{bias, weight}
backbone.fpn_lateral3.{bias, weight}
backbone.fpn_lateral4.{bias, weight}
backbone.fpn_lateral5.{bias, weight}
backbone.fpn_output2.{bias, weight}
backbone.fpn_output3.{bias, weight}
backbone.fpn_output4.{bias, weight}
backbone.fpn_output5.{bias, weight}
ins_head.cate_pred.{bias, weight}
ins_head.cate_tower.0.weight
ins_head.cate_tower.2.weight
ins_head.kernel_pred.{bias, weight}
ins_head.kernel_tower.0.weight
ins_head.kernel_tower.2.weight
mask_head.conv_pred.0.weight
mask_head.conv_pred.1.{bias, weight}
mask_head.convs_all_levels.0.conv0.0.weight
mask_head.convs_all_levels.1.conv0.0.weight
mask_head.convs_all_levels.2.conv0.0.weight
mask_head.convs_all_levels.2.conv1.0.weight
mask_head.convs_all_levels.3.conv0.0.weight
mask_head.convs_all_levels.3.conv1.0.weight
mask_head.convs_all_levels.3.conv2.0.weight
WARNING [04/29 13:43:54 fvcore.common.checkpoint]: The checkpoint state_dict contains keys that are not used by the model:
fc1000.{bias, weight}
[04/29 13:43:54 adet.trainer]: Starting training from iteration 0
/usr/local/lib/python3.8/dist-packages/detectron2/structures/image_list.py:88: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
max_size = (max_size + (stride - 1)) // stride * stride
/usr/local/lib/python3.8/dist-packages/torch/nn/functional.py:3631: 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(
/usr/local/lib/python3.8/dist-packages/torch/nn/functional.py:3679: 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(
/usr/local/lib/python3.8/dist-packages/torch/functional.py:445: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at ../aten/src/ATen/native/TensorShape.cpp:2157.)
return _VF.meshgrid(tensors, **kwargs) # type: ignore[attr-defined]
/home/xss/Projects/wood/solov2_d2/adet/modeling/solov2/solov2.py:279: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
(center_w / upsampled_size[1]) // (1. / num_grid))
/home/xss/Projects/wood/solov2_d2/adet/modeling/solov2/solov2.py:281: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
(center_h / upsampled_size[0]) // (1. / num_grid))
/home/xss/Projects/wood/solov2_d2/adet/modeling/solov2/solov2.py:285: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
0, int(((center_h - half_h) / upsampled_size[0]) // (1. / num_grid)))
/home/xss/Projects/wood/solov2_d2/adet/modeling/solov2/solov2.py:287: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
num_grid - 1, int(((center_h + half_h) / upsampled_size[0]) // (1. / num_grid)))
/home/xss/Projects/wood/solov2_d2/adet/modeling/solov2/solov2.py:289: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
0, int(((center_w - half_w) / upsampled_size[1]) // (1. / num_grid)))
/home/xss/Projects/wood/solov2_d2/adet/modeling/solov2/solov2.py:291: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
num_grid - 1, int(((center_w + half_w) / upsampled_size[1]) // (1. / num_grid)))
Traceback (most recent call last):
File "/home/xss/Projects/wood/solov2_d2/tools/train_wood.py", line 250, in
launch(
File "/usr/local/lib/python3.8/dist-packages/detectron2/engine/launch.py", line 82, in launch
main_func(*args)
File "/home/xss/Projects/wood/solov2_d2/tools/train_wood.py", line 244, in main
return trainer.train()
File "/home/xss/Projects/wood/solov2_d2/tools/train_wood.py", line 124, in train
self.train_loop(self.start_iter, self.max_iter)
File "/home/xss/Projects/wood/solov2_d2/tools/train_wood.py", line 113, in train_loop
self.run_step()
File "/usr/local/lib/python3.8/dist-packages/detectron2/engine/defaults.py", line 494, in run_step
self._trainer.run_step()
File "/usr/local/lib/python3.8/dist-packages/detectron2/engine/train_loop.py", line 285, in run_step
losses.backward()
File "/usr/local/lib/python3.8/dist-packages/torch/_tensor.py", line 307, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
File "/usr/local/lib/python3.8/dist-packages/torch/autograd/init.py", line 154, in backward
Variable._execution_engine.run_backward(
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [4, 128, 184, 232]], which is output 0 of ReluBackward0, is at version 3; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

Process finished with exit code 1

@刘看山 请问要怎么彻底卸载原版的solov2呀