Toybrick

标题: pytorch 转换时报错 [打印本页]

作者: JasonZhu 时间: 2020-3-23 19:37
标题: pytorch 转换时报错
rknn 1.3  pytorch 报错，如下：
--> Building model
E Catch exception when building RKNN model!
E Traceback (most recent call last):
E File "/root/miniconda2/envs/rknn-1.3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1626, in _create_c_op
E    c_op = c_api.TF_FinishOperation(op_desc)
E tensorflow.python.framework.errors_impl.InvalidArgumentError: Dimensions must be equal, but are 2 and 8 for 'add_at_input134.1_365/Add' (op: 'Add') with input shapes: [1,1,2,128], [1,4,8,128].
E During handling of the above exception, another exception occurred:
E Traceback (most recent call last):
E File "rknn/api/rknn_base.py", line 737, in rknn.api.rknn_base.RKNNBase.build
E File "rknn/api/rknn_base.py", line 1644, in rknn.api.rknn_base.RKNNBase._quantize2
E File "rknn/base/RKNNlib/app/medusa/quantization.py", line 105, in rknn.base.RKNNlib.app.medusa.quantization.Quantization.run
E File "rknn/base/RKNNlib/app/medusa/quantization.py", line 44, in rknn.base.RKNNlib.app.medusa.quantization.Quantization._run_quantization
E File "rknn/base/RKNNlib/app/medusa/workspace.py", line 129, in rknn.base.RKNNlib.app.medusa.workspace.Workspace.run
E File "rknn/base/RKNNlib/app/medusa/workspace.py", line 99, in rknn.base.RKNNlib.app.medusa.workspace.Workspace._setup_graph
E File "rknn/base/RKNNlib/app/medusa/workspace.py", line 100, in rknn.base.RKNNlib.app.medusa.workspace.Workspace._setup_graph
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 274, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 278, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 305, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build_layer
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 305, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build_layer
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 305, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build_layer
E [Previous line repeated 124 more times]
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 331, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build_layer
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 336, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build_layer
E File "rknn/base/RKNNlib/layer/RKNNlayer.py", line 287, in rknn.base.RKNNlib.layer.RKNNlayer.RKNNLayer.compute_tensor
E File "rknn/base/RKNNlib/layer/add.py", line 36, in rknn.base.RKNNlib.layer.add.Add.compute_out_tensor
E File "/root/miniconda2/envs/rknn-1.3/lib/python3.6/site-packages/tensorflow/python/ops/gen_math_ops.py", line 301, in add
E    "Add", x=x, y=y, name=name)
E File "/root/miniconda2/envs/rknn-1.3/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
E    op_def=op_def)
E File "/root/miniconda2/envs/rknn-1.3/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py", line 488, in new_func
E    return func(*args, **kwargs)
E File "/root/miniconda2/envs/rknn-1.3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3272, in create_op
E    op_def=op_def)
E File "/root/miniconda2/envs/rknn-1.3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1790, in __init__
E    control_input_ops)
E File "/root/miniconda2/envs/rknn-1.3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1629, in _create_c_op
E    raise ValueError(str(e))
E ValueError: Dimensions must be equal, but are 2 and 8 for 'add_at_input134.1_365/Add' (op: 'Add') with input shapes: [1,1,2,128], [1,4,8,128].
Build pytorch failed!

作者: JasonZhu 时间: 2020-3-23 19:39
#########################################################################
##
## Some utility for training, data processing, and network.
##
#########################################################################
import torch
import torch.nn as nn
from parameters import Parameters

p = Parameters()

def backward_hook(self, grad_input, grad_output):
print('grad_input norm:', grad_input[0].data.norm())

def cross_entropy2d(inputs, target, weight=None, size_average=True):
loss = torch.nn.CrossEntropyLoss()

n, c, h, w = inputs.size()
prediction = inputs.transpose(1, 2).transpose(2, 3).contiguous().view(-1, c)
gt =target.transpose(1, 2).transpose(2, 3).contiguous().view(-1)

return loss(prediction, gt)

######################################################################
##
## Convolution layer modules
##
######################################################################
class Conv2D_BatchNorm_Relu(nn.Module):
def __init__(self, in_channels, n_filters, k_size, padding, stride, bias=True, acti=True):
      super(Conv2D_BatchNorm_Relu, self).__init__()

      if acti:
         self.cbr_unit = nn.Sequential(nn.Conv2d(in_channels, n_filters, k_size,
                                                padding=padding, stride=stride, bias=bias),
                                 nn.BatchNorm2d(n_filters),
                                 nn.ReLU(inplace=True),)
      else:
         self.cbr_unit = nn.Conv2d(in_channels, n_filters, k_size, padding=padding, stride=stride, bias=bias)

def forward(self, inputs):
      outputs = self.cbr_unit(inputs)
      return outputs

class bottleneck(nn.Module):
def __init__(self, in_channels, out_channels, acti=True):
      super(bottleneck, self).__init__()
      self.acti = acti
      temp_channels = in_channels//4
      if in_channels < 4:
         temp_channels = in_channels
      self.conv1 = Conv2D_BatchNorm_Relu(in_channels, temp_channels, 1, 0, 1)
      self.conv2 = Conv2D_BatchNorm_Relu(temp_channels, temp_channels, 3, 1, 1)
      self.conv3 = Conv2D_BatchNorm_Relu(temp_channels, out_channels, 1, 0, 1, acti = self.acti)

      self.residual = Conv2D_BatchNorm_Relu(in_channels, out_channels, 1, 0, 1)

def forward(self, x):
      re = x

      out = self.conv1(x)
      out = self.conv2(out)
      out = self.conv3(out)
      if not self.acti:
         return out

      re = self.residual(x)
      out = out + re

      return out

class bottleneck_down(nn.Module):
def __init__(self, in_channels, out_channels):
      super(bottleneck_down, self).__init__()
      temp_channels = in_channels//4
      if in_channels < 4:
         temp_channels = in_channels
      self.conv1 = Conv2D_BatchNorm_Relu(in_channels, temp_channels, 1, 0, 1)
      self.conv2 = Conv2D_BatchNorm_Relu(temp_channels, temp_channels, 3, 1, 2)
      self.conv3 = Conv2D_BatchNorm_Relu(temp_channels, out_channels, 1, 0, 1)

      self.residual = Conv2D_BatchNorm_Relu(in_channels, out_channels, 3, 1, 2)

def forward(self, x):
      re = x

      out = self.conv1(x)
      out = self.conv2(out)
      out = self.conv3(out)

      re = self.residual(x)

      out = out + re

      return out

class bottleneck_up(nn.Module):
def __init__(self, in_channels, out_channels):
      super(bottleneck_up, self).__init__()
      temp_channels = in_channels//4
      if in_channels < 4:
         temp_channels = in_channels
      self.conv1 = Conv2D_BatchNorm_Relu(in_channels, temp_channels,1,  0, 1)
      self.conv2 = nn.Sequential( nn.ConvTranspose2d(temp_channels, temp_channels, 3, 2, 1, 1),
                                    nn.BatchNorm2d(temp_channels),
                                    nn.ReLU() )
      self.conv3 = Conv2D_BatchNorm_Relu(temp_channels, out_channels, 1, 0, 1)

      self.residual = nn.Sequential( nn.ConvTranspose2d(in_channels, out_channels, 3, 2, 1, 1),
                                    nn.BatchNorm2d(out_channels),
                                    nn.ReLU() )

def forward(self, x):
      re = x

      out = self.conv1(x)
      out = self.conv2(out)
      out = self.conv3(out)

      re = self.residual(re)

      out = out + re

      return out

class Output(nn.Module):
def __init__(self, in_size, out_size):
      super(Output, self).__init__()
      self.conv = bottleneck(in_size, out_size, acti=False)

def forward(self, inputs):
      outputs = self.conv(inputs)
      return outputs

class hourglass_same(nn.Module):
def __init__(self, in_channels, out_channels):
      super(hourglass_same, self).__init__()
      self.down1 = bottleneck_down(in_channels, out_channels)
      self.down2 = bottleneck_down(out_channels, out_channels)
      self.down3 = bottleneck_down(out_channels, out_channels)
      self.down4 = bottleneck_down(out_channels, out_channels)

      self.same1 = bottleneck(out_channels, out_channels)
      self.same2 = bottleneck(out_channels, out_channels)

      self.up2 = bottleneck_up(out_channels, out_channels)
      self.up3 = bottleneck_up(out_channels, out_channels)
      self.up4 = bottleneck_up(out_channels, out_channels)
      self.up5 = bottleneck_up(out_channels, out_channels)

      self.residual1 = bottleneck_down(in_channels, out_channels)
      self.residual2 = bottleneck_down(out_channels, out_channels)
      self.residual3 = bottleneck_down(out_channels, out_channels)
      self.residual4 = bottleneck_down(out_channels, out_channels)

def forward(self, inputs):
      outputs1 = self.down1(inputs)  # 512*256 -> 256*128
      outputs2 = self.down2(outputs1)  # 256*128 -> 128*64
      outputs3 = self.down3(outputs2)  # 128*64 -> 64*32
      outputs4 = self.down4(outputs3)  # 64*32 -> 32*16

      outputs = self.same1(outputs4)  # 16*8 -> 16*8
      outputs = self.same2(outputs)  # 16*8 -> 16*8

      outputs = self.up2(outputs + self.residual4(outputs3))  # 32*16 -> 64*32
      outputs = self.up3(outputs + self.residual3(outputs2))  # 64*32 -> 128*64
      outputs = self.up4(outputs + self.residual2(outputs1))  # 128*64 -> 256*128
      outputs = self.up5(outputs + self.residual1(inputs))  # 256*128 -> 512*256

      return outputs

class resize_layer(nn.Module):
def __init__(self, in_channels, out_channels, acti = True):
      super(resize_layer, self).__init__()
      self.conv = Conv2D_BatchNorm_Relu(in_channels, out_channels//2, 7, 3, 2)
      self.maxpool = nn.MaxPool2d(2, 2)
      self.re1 = bottleneck(out_channels//2, out_channels//2)
      self.re2 = bottleneck(out_channels//2, out_channels//2)
      self.re3 = bottleneck(out_channels//2, out_channels)

def forward(self, inputs):
      outputs = self.conv(inputs)
      outputs = self.re1(outputs)
      outputs = self.maxpool(outputs)
      outputs = self.re2(outputs)
      outputs = self.maxpool(outputs)
      outputs = self.re3(outputs)

      return outputs

class hourglass_block(nn.Module):
def __init__(self, in_channels, out_channels, acti = True, input_re=True):
      super(hourglass_block, self).__init__()
      self.layer1 = hourglass_same(in_channels, out_channels)
      self.re1 = bottleneck(out_channels, out_channels)
      self.re2 = bottleneck(out_channels, out_channels)
      self.re3 = bottleneck(1, out_channels)

      self.out_confidence = Output(out_channels, 1)
      self.out_offset = Output(out_channels, 2)
      self.out_instance = Output(out_channels, p.feature_size)
      self.input_re = input_re

def forward(self, inputs):
      outputs = self.layer1(inputs)
      outputs = self.re1(outputs)

      out_confidence = self.out_confidence(outputs)
      out_offset = self.out_offset(outputs)
      out_instance = self.out_instance(outputs)

      out = out_confidence

      outputs = self.re2(outputs)
      out = self.re3(out)

      if self.input_re:
         outputs = outputs + out + inputs
      else:
         outputs = outputs + out

      return [out_confidence, out_offset, out_instance], outputs

作者: jefferyzhang 时间: 2020-3-24 08:38
E File "/root/miniconda2/envs/rknn-1.3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1629, in _create_c_op
E raise ValueError(str(e))

这个报错都错在tensorflow里，与rknn无关的。先检查你自己的pb文件能不能被当前这个版本tensorflow读出来并推理。

作者: JasonZhu 时间: 2020-3-24 17:49

jefferyzhang 发表于 2020-3-24 08:38
E File "/root/miniconda2/envs/rknn-1.3/lib/python3.6/site-packages/tensorflow/python/framework/ops ...

我的模型是pytorch的，可以用tensorflow 加载？

作者: jefferyzhang 时间: 2020-3-24 17:53

JasonZhu 发表于 2020-3-24 17:49
我的模型是pytorch的，可以用tensorflow 加载？

那这就尴尬了，请把模型和转换脚本一起发给我下，我报给NPU部门debug

作者: JasonZhu 时间: 2020-3-24 18:00

jefferyzhang 发表于 2020-3-24 17:53
那这就尴尬了，请把模型和转换脚本一起发给我下，我报给NPU部门debug

我现在redmine上提上去看看，多谢版主

作者: jefferyzhang 时间: 2020-3-25 08:17

JasonZhu 发表于 2020-3-24 18:00
我现在redmine上提上去看看，多谢版主

嗯，好的

作者: Sean 时间: 2020-3-25 09:13
也遇到了同样问题，楼主能加个ｗｘ聊聊吗？

--> Building model
E Catch exception when building RKNN model!
E Traceback (most recent call last):
E File "/home/user/.local/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1567, in _create_c_op
E    c_op = c_api.TF_FinishOperation(op_desc)
E tensorflow.python.framework.errors_impl.InvalidArgumentError: Dimension 1 in both shapes must be equal, but are 6 and 26. Shapes are [1,6,6] and [1,26,26]. for 'cat_at_2304_1/cat_at_2304_1' (op: 'ConcatV2') with input shapes: [1,6,6,256], [1,26,26,512], [] and with computed input tensors: input[2] = <3>.
E During handling of the above exception, another exception occurred:
E Traceback (most recent call last):
E File "rknn/api/rknn_base.py", line 737, in rknn.api.rknn_base.RKNNBase.build
E File "rknn/api/rknn_base.py", line 1644, in rknn.api.rknn_base.RKNNBase._quantize2
E File "rknn/base/RKNNlib/app/medusa/quantization.py", line 105, in rknn.base.RKNNlib.app.medusa.quantization.Quantization.run
E File "rknn/base/RKNNlib/app/medusa/quantization.py", line 44, in rknn.base.RKNNlib.app.medusa.quantization.Quantization._run_quantization
E File "rknn/base/RKNNlib/app/medusa/workspace.py", line 129, in rknn.base.RKNNlib.app.medusa.workspace.Workspace.run
E File "rknn/base/RKNNlib/app/medusa/workspace.py", line 99, in rknn.base.RKNNlib.app.medusa.workspace.Workspace._setup_graph
E File "rknn/base/RKNNlib/app/medusa/workspace.py", line 100, in rknn.base.RKNNlib.app.medusa.workspace.Workspace._setup_graph
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 274, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 278, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 305, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build_layer
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 331, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build_layer
E File "rknn/base/RKNNlib/RKNNnetbuilder.py", line 336, in rknn.base.RKNNlib.RKNNnetbuilder.RKNNNetBuilder.build_layer
E File "rknn/base/RKNNlib/layer/RKNNlayer.py", line 287, in rknn.base.RKNNlib.layer.RKNNlayer.RKNNLayer.compute_tensor
E File "rknn/base/RKNNlib/layer/concat.py", line 44, in rknn.base.RKNNlib.layer.concat.Concat.compute_out_tensor
E File "/home/user/.local/lib/python3.6/site-packages/tensorflow/python/ops/array_ops.py", line 1189, in concat
E    return gen_array_ops.concat_v2(values=values, axis=axis, name=name)
E File "/home/user/.local/lib/python3.6/site-packages/tensorflow/python/ops/gen_array_ops.py", line 953, in concat_v2
E    "ConcatV2", values=values, axis=axis, name=name)
E File "/home/user/.local/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
E    op_def=op_def)
E File "/home/user/.local/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3392, in create_op
E    op_def=op_def)
E File "/home/user/.local/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1734, in __init__
E    control_input_ops)
E File "/home/user/.local/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1570, in _create_c_op
E    raise ValueError(str(e))
E ValueError: Dimension 1 in both shapes must be equal, but are 6 and 26. Shapes are [1,6,6] and [1,26,26]. for 'cat_at_2304_1/cat_at_2304_1' (op: 'ConcatV2') with input shapes: [1,6,6,256], [1,26,26,512], [] and with computed input tensors: input[2] = <3>.
Build pytorch failed!

作者: Sean 时间: 2020-3-25 09:14

Sean 发表于 2020-3-25 09:13
也遇到了同样问题，楼主能加个ｗｘ聊聊吗？

--> Building model

用的是torch的yolo3代码

作者: wali 时间: 2020-4-7 20:35
请问这个问题有解决方案了么？遇到了同样的问题

欢迎光临 Toybrick (https://t.rock-chips.com/)