caffemodel 当blob只有shape没有data时会有异常

[tian161]

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

使用X2Paddle 进行 caffe转nb，当模型文件caffemodel的blob权重参数中只有shape没有data时，会导致报错。
这是我的模型文件.caffemodel的前面几层
name: ""
layer {
name: "input_1"
type: "Input"
top: "input_1"
phase: TEST
input_param {
shape {
dim: 1
dim: 3
dim: 160
dim: 40
}
}
}
layer {
name: "conv0"
type: "Convolution"
bottom: "input_1"
top: "conv0"
blobs {
shape {
dim: 32
dim: 3
dim: 3
dim: 3
}
}
blobs {
shape {
dim: 32
}
}
phase: TEST
convolution_param {
num_output: 32
bias_term: true
pad_h: 1
pad_w: 1
kernel_h: 3
kernel_w: 3
stride_h: 1
stride_w: 1
}
}
layer {
name: "bn0"
type: "BatchNorm"
bottom: "conv0"
top: "bn0"
blobs {
data: -0.007865031249821186
data: -0.005324164871126413
data: -0.010764184407889843
data: 0.008240368217229843
data: 0.01976969465613365
data: 0.012027844786643982
data: 0.002962957601994276
data: -0.3145563304424286
data: 0.10758233815431595
data: 0.025216389447450638
data: -0.07927775382995605
data: -0.004488849081099033
data: -0.009283231571316719
data: -0.020594490692019463
data: -0.05032395198941231
data: 0.14864583313465118
data: -0.03062831051647663
data: -0.027901044115424156
data: -0.007554500363767147
data: -0.05905086174607277
data: 0.013064151629805565
data: -0.001625223783776164
data: -0.012637660838663578
data: -0.020215753465890884
data: -0.15577200055122375
data: -0.005744390655308962
data: -0.004265745636075735
data: 0.020425138995051384
data: -0.0014742618659511209
data: 0.02395986020565033
data: 0.020134909078478813
data: -0.013855302706360817
shape {
dim: 32
}
}
blobs {
data: 0.012731957249343395
data: 0.008646254427731037
data: 0.011110151186585426
data: 0.009516902267932892
data: 0.008103199303150177
data: 0.008265161886811256
data: 0.00312191154807806
data: 0.02370261959731579
data: 0.0027517329435795546
data: 0.013189432211220264
data: 0.007192822638899088
data: 0.008816424757242203
data: 0.007724612019956112
data: 0.014152021147310734
data: 0.01284481305629015
data: 0.013880057260394096
data: 0.014332244172692299
data: 0.0011996557004749775
data: 0.015406877733767033
data: 0.010128257796168327
data: 0.002697761869058013
data: 0.0050042783841490746
data: 0.011377030052244663
data: 0.00816092174500227
data: 0.010117243975400925
data: 0.011570217087864876
data: 0.005011581815779209
data: 0.0009002541191875935
data: 0.0012184175429865718
data: 0.016399741172790527
data: 0.006782908923923969
data: 0.004564266186207533
shape {
dim: 32
}
}
blobs {
data: 1.0
shape {
dim: 1
}
}
phase: TEST
batch_norm_param {
moving_average_fraction: 0.9900000095367432
eps: 0.0010000000474974513
}
}

• 错误信息

• 错误截图

  File "F:\PyProject\PaddleOPT\Main.py", line 5, in
  caffe2paddle(
  File "F:\PyProject\PaddleOPT\X2Paddle\x2paddle\convert.py", line 240, in caffe2paddle
  model = CaffeDecoder(proto_file, weight_file, caffe_proto)
  File "f:\PyProject\PaddleOPT\lib\site-packages\x2paddle-1.4.1-py3.10.egg\x2paddle\decoder\caffe_decoder.py", line 292, in init
  self.load_using_pb()
  File "f:\PyProject\PaddleOPT\lib\site-packages\x2paddle-1.4.1-py3.10.egg\x2paddle\decoder\caffe_decoder.py", line 306, in load_using_pb
  self.params = [pair(layer) for layer in layers if layer.blobs]
  File "f:\PyProject\PaddleOPT\lib\site-packages\x2paddle-1.4.1-py3.10.egg\x2paddle\decoder\caffe_decoder.py", line 306, in
  self.params = [pair(layer) for layer in layers if layer.blobs]
  File "f:\PyProject\PaddleOPT\lib\site-packages\x2paddle-1.4.1-py3.10.egg\x2paddle\decoder\caffe_decoder.py", line 305, in
  pair = lambda layer: (layer.name, self.normalize_pb_data(layer))
  File "f:\PyProject\PaddleOPT\lib\site-packages\x2paddle-1.4.1-py3.10.egg\x2paddle\decoder\caffe_decoder.py", line 330, in normalize_pb_data
  list(blob.data), dtype=np.float32).reshape(c_o, c_i, h, w)
  ValueError: cannot reshape array of size 0 into shape (32,3,3,3)

具体信息

• 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：https://github.com/Megvii-BaseDetection/YOLOX
• 应用场景
  用于移动端业务下的检测业务
• 版本信息
  PaddlePaddle => ❔ 2.4.2.post117
  X2Paddle => ❔ 1.4.1
  来源框架版本(PyTorch/TF/ONNX/Caffe) => ❔ Caffe
• 您的联系方式(邮箱/微信/电话)
  18678297327