在自己的数据集上微调Light CNN

最新推荐文章于 2025-07-01 20:14:10 发布

原创

最新推荐文章于 2025-07-01 20:14:10 发布 · 3.3k 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#深度学习-人脸识别

本文档记录了在自定义数据集上微调Light CNN模型的过程中遇到的问题，包括数据预处理（灰度化）、网络结构定义、训练集与验证集划分以及训练过程中的错误。同时，作者寻求关于解决报错和模型压缩的方法。

遇到的坑

light cnn 加载的是灰度图，所以在 data 层要添加参数：

is_color:false
示例如下：

layer {
  name: "data"
  type: "ImageData"
  top: "data"
  top: "label"
  image_data_param{
      source: "/home/code/caffe-master/lightCNNFace/val.txt"
      batch_size: 20
      is_color:false
    }
  transform_param {
    scale: 0.00390625
    crop_size: 128
    mirror: false
  }
  include: { phase: TEST }
}

1.编写微调的网络结构lightcnn_train_test.prototxt

2.准备训练图像数据，分成训练集和验证集合

注意要将图像灰度化

3.运行训练脚本文件

code/caffe-master/build/tools/caffe train \
    --solver=code/caffe-master/lightCNNFace/LCNN_solver.prototxt \
    --weights=code/caffe-master/lightCNNFace/LightenedCNN_B.caffemodel -gpu=0

4.微调时报错

  Cannot copy param 0 weights from layer 'conv1'; shape mismatch.  Source param shape is 96 1 5 5 (2400); target param shape is 96 3 5 5 (7200). To learn this layer's parameters from scratch rather than copying from a saved net, rename the layer

5.附 lightcnn_train_test.prototxt

有过微调light cnn的网友吗？求解决方案，另外如何进一步减小模型大小？

name: "DeepFace_set003_net"

layer {
  name: "data"
  type:"ImageData"
  top: "data"
  top: "label"
  image_data_param{
      source: "/home/himon/code/caffe-master/lightCNNFace/train.txt"
      batch_size: 20
      shuffle: true
    }
  transform_param {
    scale: 0.00390625
    crop_size: 128
    mirror: true

  }
  include: { phase: TRAIN }
}

layer {
  name: "data"
  type: "ImageData"
  top: "data"
  top: "label"
  image_data_param{
      source: "/home/himon/code/caffe-master/lightCNNFace/val.txt"
      batch_size: 20
    }
  transform_param {
    scale: 0.00390625
    crop_size: 128
    mirror: false
  }
  include: { phase: TEST }
}

layer{
  name: "conv1"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 5
    stride: 1
    pad: 2
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0.1
    }
  }
  bottom: "data"
  top: "conv1"
}

layer{
  name: "slice1"
  type:"Slice"
  slice_param {
    slice_dim: