yolov5 onnx 转 infer 报错 Scale's shape must be 2 or 1, but got shape = 0

[maiquanshen]

感谢您参与 X2Paddle 社区！ 问题模版为了 X2Paddle 能更好的迭代，例如新功能发布、 RoadMaps 和错误跟踪. 😸

问题描述

• 错误信息
• (paddle_env) iotmp@iotmp-PR4024ZY:/media/sdb3/PaddleOCR$ x2paddle --framework=onnx --model=/media/sdb3/rknn_model_zoo/models/vision/object_detection/yolov5-pytorch/yolov5/runs/capchat_20241202/exp4/weights/best.onnx --save_dir=./capchat/capchat_20241203_infer --enable_onnx_checker=true --input_shape_dict="{'image':[1, 3, 640, 640]}"
  /home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/setuptools/command/easy_install.py:41: DeprecationWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html
  import pkg_resources
  /home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/pkg_resources/init.py:3154: DeprecationWarning: Deprecated call to pkg_resources.declare_namespace('mpl_toolkits').
  Implementing implicit namespace packages (as specified in PEP 420) is preferred to pkg_resources.declare_namespace. See https://setuptools.pypa.io/en/latest/references/keywords.html#keyword-namespace-packages
  declare_namespace(pkg)
  /home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/pkg_resources/init.py:3154: DeprecationWarning: Deprecated call to pkg_resources.declare_namespace('google').
  Implementing implicit namespace packages (as specified in PEP 420) is preferred to pkg_resources.declare_namespace. See https://setuptools.pypa.io/en/latest/references/keywords.html#keyword-namespace-packages
  declare_namespace(pkg)
  INFO:root:paddle.version = 2.4.0
  INFO:root:Now translating model from onnx to paddle.
  model ir_version: 4, op version: 9
  Shape inferencing ...
  Stopping at incomplete shape inference at Concat: x2paddle__model_12_Concat_output_0
  node inputs:
  name: "x2paddle__model_11_Upsample_output_0"
  type {
  tensor_type {
  elem_type: 1
  shape {
  dim {
  dim_param: "unk__0"
  }
  dim {
  dim_param: "unk__1"
  }
  dim {
  dim_param: "unk__2"
  }
  dim {
  dim_param: "unk__3"
  }
  }
  }
  }

name: "x2paddle__model_6_cv3_act_Relu_output_0"
type {
tensor_type {
elem_type: 1
shape {
dim {
dim_value: 1
}
dim {
dim_value: 256
}
dim {
dim_value: 40
}
dim {
dim_value: 40
}
}
}
}

node outputs:
name: "x2paddle__model_12_Concat_output_0"
type {
tensor_type {
elem_type: 1
shape {
dim {
}
dim {
dim_param: "unk__1 + 256"
}
dim {
}
dim {
}
}
}
}

!!!!!!!!!!
Shape inferenced.
/home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/x2paddle/decoder/onnx_decoder.py:374: DeprecationWarning: mapping.TENSOR_TYPE_TO_NP_TYPE is now deprecated and will be removed in a future release.To silence this warning, please use helper.tensor_dtype_to_np_dtype instead.
TENSOR_TYPE_TO_NP_TYPE[item.type.tensor_type.elem_type],
Now, onnx2paddle support convert onnx model opset_verison [7, 8, 9, 10, 11, 12, 13, 14, 15], opset_verison of your onnx model is 9.
Total nodes: 150
Nodes converting ...
Converting node 1 ... /home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/x2paddle/decoder/onnx_decoder.py:172: DeprecationWarning: mapping.TENSOR_TYPE_TO_NP_TYPE is now deprecated and will be removed in a future release.To silence this warning, please use helper.tensor_dtype_to_np_dtype instead.
return TENSOR_TYPE_TO_NP_TYPE[dtype]
Converting node 2 ... /home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/x2paddle/decoder/onnx_decoder.py:175: DeprecationWarning: mapping.TENSOR_TYPE_TO_NP_TYPE is now deprecated and will be removed in a future release.To silence this warning, please use helper.tensor_dtype_to_np_dtype instead.
return TENSOR_TYPE_TO_NP_TYPE[dtype]
Converting node 276 ...
Nodes converted.
INFO:root:Model optimizing ...
INFO:root:Model optimized.
Exporting inference model from python code ('/media/sdb3/PaddleOCR/capchat/capchat_20241203_infer/x2paddle_code.py')...

In transformed code:

File "/media/sdb3/PaddleOCR/./capchat/capchat_20241203_infer/x2paddle_code.py", line 247, in forward
    x2paddle__model_14_act_Relu_output_0 = self.relu39(x2paddle__model_14_conv_Conv_output_0)
    x2paddle__model_11_Constant_output_0 = paddle.slice(input=x2paddle__model_11_Constant_output_0, axes=[0], starts=[2], ends=[4])
    x2paddle__model_15_Upsample_output_0 = paddle.nn.functional.interpolate(x=x2paddle__model_14_act_Relu_output_0, scale_factor=x2paddle__model_11_Constant_output_0, mode='nearest', align_mode=1)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
    x2paddle__model_16_Concat_output_0 = paddle.concat(x=[x2paddle__model_15_Upsample_output_0, x2paddle__model_4_cv3_act_Relu_output_0], axis=1)
    x2paddle__model_17_cv1_conv_Conv_output_0 = self.conv44(x2paddle__model_16_Concat_output_0)

File "/home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/paddle/nn/functional/common.py", line 698, in interpolate
    helper.append_op(
File "/home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/paddle/fluid/layer_helper.py", line 45, in append_op
    return self.main_program.current_block().append_op(*args, **kwargs)
File "/home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/paddle/fluid/framework.py", line 4017, in append_op
    op = Operator(
File "/home/iotmp/anaconda3/envs/paddle_env/lib/python3.8/site-packages/paddle/fluid/framework.py", line 3016, in __init__
    self.desc.infer_shape(self.block.desc)

ValueError: (InvalidArgument) Scale's shape must be 2 or 1, but got shape = 0 .

[Hint: Expected scale_tensor_dim[0] == 2 || scale_tensor_dim[0] == 1 == true, but received scale_tensor_dim[0] == 2 || scale_tensor_dim[0] == 1:0 != true:1.] (at /paddle/paddle/phi/infermeta/multiary.cc:1514)
[operator < nearest_interp_v2 > error]

• 错误截图
  

具体信息

onnx模型已经是去掉sacle_factor以及后处理部分，在转的是还会因为sacle_factor这个分支没有而报错，为什么呢，如何去掉呢

• 转换模型后用处
  • [ *] 使用 Paddle 框架/ PaddleInference 推理预测
  • 使用 Paddle-Lite 做移动端推理
  • 转换预训练参数，再使用 Paddle 进行模型开发
• 模型来源
  YOLOX：yolov5
• 应用场景
  边缘端的检测业务
• 版本信息
  PaddlePaddle => ❔ 2.4.2
  X2Paddle => ❔ 1.5.0
  来源框架版本(PyTorch/TF/ONNX/Caffe) => ❔
• 您的联系方式(邮箱/微信/电话)
  [email protected]

[maiquanshen]

用了一个笨的方法，自己解决了，修改生成后的x2paddle_code.py代码，将sacle_factor的输入去掉，然后手动给一个默认值。最后修改x2paddle的源码，让它用修改后的x2paddle_code去保存权重文件