×

签到

分享到微信

打开微信,使用扫一扫进入页面后,点击右上角菜单,

点击“发送给朋友”或“分享到朋友圈”完成分享

MLU220量化模型时报以下错误:ImageData or Data layer don&#039t require preprocessing 已解决 wangbencun2021-08-02 14:00:52 回复 7 查看 技术答疑 使用求助 社区交流
MLU220量化模型时报以下错误:ImageData or Data layer don&#039t require preprocessing
分享到:

操作步骤:

在MLU220上配置ini文件如下:

[model]
;below two are lists, depending on work
original_models_path = /caffe/build/tools/aa.prototxt
save_model_path = /caffe/build/tools/aa_quan.prototxt
[data]
;only one should be set for below two
images_list_path = /caffe/build/tools/img_files.txt
;images_db_path =  /path/to/lmdb
used_images_num = 4
[weights]
original_weights_path = /caffe/build/tools/a.caffemodel
[preprocess]
mean = 0,0,0
std = 0.00392156
scale = 256,192
crop = 256, 192
[config]
quantize_op_list = Conv, FC, LRN
use_firstconv = 1
;int8_ _list = conv3, conv9
;int16_ _list = conv3, conv9
[custom]
use_custom_preprocess = 0
;only support ARGB, ABGR, BGRA, RGBA
input_format = BGRA
;only support BGR, RGB
filter_format = BGR
根据提示将aa.prototxt的输入层修改为:

{
 name: "data"
 type: "ImageData"
 top: "data"
 top: "label"
 transform_param {
  mean_value: 104
  mean_value: 117
  mean_value: 123
  scale: 1
 }
 image_data_param {
  source: "./img_files.txt"
  batch_size: 1
  new_height: 256
  new_width: 192
 }
}
执行./generate_quantized_pt -ini_file convert.ini后报以下错误:

I0802 13:49:44.062245  2740 generate_quantized_pt.cpp:477] ImageData or Data don't require preprocessing!
I0802 13:49:44.063182  2740 net.cpp:1708] reshaping...
I0802 13:49:44.068333  2740 net.cpp:1738] reshaping done...
Killed

请问这个问题该如何解决呢?

版权所有 © 2022 寒武纪 Cambricon.com 备案/许可证号:京ICP备17003415号-1
关闭