目录
网上参考方法,没成功:
ok的解决方法:
1. 在yolov5s的pytorch模型转换onnx模型时报如下错误:
RuntimeError: step!=1 is currently not supported
原因主要是低版本的opset不支持切片操作导致的;
把模型转换的代码改成如下所示即可,即使用版本11以上的opset:
torch.onnx.export(model, img, "xxx.onnx", verbose=True,opset_version=11,export_params=True)
2. 解决该问题后可能会继续出现如下错误:
RuntimeError: Exporting the operator silu to ONNX opset version 11 is not supported. Please open a bug to request ONNX export support for the missing operator.
这是因为onnx与pytorch一些方法不兼容导致的,onnx不支持silu,把激活函数换一种写法即可。
解决方法如下:
pytorch的激活函数源码位置:X:\anaconda3\envs\python3.7\Lib\sitepackages\torch\nn\modules\activation.py(此处结合自己的anaconda实际安装位置来更改)。
class SiLU(Module):
__constants__ = ['inplace']
inplace: bool
def __init__(self, inplace: bool = False):
super(SiLU, self).__init__()
self.inplace = inplace
def forward(self, input: Tensor) -> Tensor:
# ------------------------------------- #
# 把F.silu替换掉,修改后如下
return input * torch.sigmoid(input)
#原来的代码
return F.silu(input, inplace=self.inplace)
yolov7导出时也报错,我改了上面也继续报错。
原文链接:https://blog.csdn.net/weixin_37889356/article/details/121671974
class SiLU(nn.Module):
"""export-friendly version of nn.SiLU()"""
@staticmethod
def forward(x):
return x * torch.sigmoid(x)
def replace_module(module, replaced_module_type, new_module_type, replace_func=None) -> nn.Module:
"""
Replace given type in module to a new type. mostly used in deploy.
Args:
module (nn.Module): model to apply replace operation.
replaced_module_type (Type): module type to be replaced.
new_module_type (Type)
replace_func (function): python function to describe replace logic. Defalut value None.
Returns:
model (nn.Module): module that already been replaced.
"""
def default_replace_func(replaced_module_type, new_module_type):
return new_module_type()
if replace_func is None:
replace_func = default_replace_func
model = module
if isinstance(module, replaced_module_type):
model = replace_func(replaced_module_type, new_module_type)
else: # recurrsively replace
for name, child in module.named_children():
new_child = replace_module(child, replaced_module_type, new_module_type)
if new_child is not child: # child is already replaced
model.add_module(name, new_child)
return model
model = replace_module(model, nn.SiLU, SiLU)
model = replace_module(model, nn.Hardsigmoid, HardSigmoid)
来自:
PPYOLOE_pytorch/model_utils.py at 545b6a68ef5765769f693dedb4cd0e086ad4eea2 · Nioolek/PPYOLOE_pytorch · GitHub