- 微信
- 微博
  
  分享文章到微博
- 复制链接
  
  复制链接到剪贴板

不同检测模型预处理中的归一化操作对比 | YOLO 系列 | 【归一化操作归纳整理】

墨理学AI 发表于 2022/01/11 22:13:24 2022/01/11

【摘要】文章目录 YOLO 系列论文不同【检测模型】预处理中的归一化操作对比YOLO系列网络结构YOLOYolov2Yolov3Yolov4Yolov5 YOLO 系列论文 ...

文章目录

YOLO 系列论文

论文地址

YOLOv4: Optimal Speed and Accuracy of Object Detection

论文翻译

YOLOv4 全文中文翻译
YOLOv3 翻译

论文解析

YOLOv4 详解版！一句话总结：速度差不多的精度碾压，精度差不多的速度碾压！

代码

总结

YOLO YOLOv2 YOLOv3 YOLOv4 都首先基于 darknet 实现
然而，当前热度较高的版本通常还是 pytorch 实现，例如 yolov5

不同【检测模型】预处理中的归一化操作对比

# 如下归一化的模型有：

yolov2  

    const float mean_vals[3] = {1.0f, 1.0f, 1.0f};
    const float norm_vals[3] = {0.007843f, 0.007843f, 0.007843f};

yolov3

    const float mean_vals[3] = {127.5f, 127.5f, 127.5f};
    const float norm_vals[3] = {0.007843f, 0.007843f, 0.007843f};

yolov4

    const float mean_vals[3] = {0, 0, 0};
    const float norm_vals[3] = {1 / 255.f, 1 / 255.f, 1 / 255.f};
    in.substract_mean_normalize(mean_vals, norm_vals);

yolov5

    const float norm_vals[3] = {1 / 255.f, 1 / 255.f, 1 / 255.f};
    in_pad.substract_mean_normalize(0, norm_vals);

squeezenetssd squeezenet  squeezenet_c_api

    const float mean_vals[3] = {104.f, 117.f, 123.f};
    in.substract_mean_normalize(mean_vals, 0);
simplepose

    const float mean_vals[3] = {0.485f * 255.f, 0.456f * 255.f, 0.406f * 255.f};
    const float norm_vals[3] = {1 / 0.229f / 255.f, 1 / 0.224f / 255.f, 1 / 0.225f / 255.f};
    
shufflenetv2  
   const float norm_vals[3] = {1 / 255.f, 1 / 255.f, 1 / 255.f};
    in.substract_mean_normalize(0, norm_vals);

scrfd

    const float mean_vals[3] = {127.5f, 127.5f, 127.5f};
    const float norm_vals[3] = {1 / 128.f, 1 / 128.f, 1 / 128.f};
rfcn  fasterrcnn

    const float mean_vals[3] = {102.9801f, 115.9465f, 122.7717f};
    in.substract_mean_normalize(mean_vals, 0);

nanodet
    const float mean_vals[3] = {103.53f, 116.28f, 123.675f};
    const float norm_vals[3] = {0.017429f, 0.017507f, 0.017125f};

mobilenetv3ssdlite

    const float mean_vals[3] = {123.675f, 116.28f, 103.53f};
    const float norm_vals[3] = {1.0f, 1.0f, 1.0f};
    
mobilenetv2ssdlite

    const float mean_vals[3] = {127.5f, 127.5f, 127.5f};
    const float norm_vals[3] = {1.0 / 127.5, 1.0 / 127.5, 1.0 / 127.5};
  
 
  1
  2
  3
  4
  5
  6
  7
  8
  9
  10
  11
  12
  13
  14
  15
  16
  17
  18
  19
  20
  21
  22
  23
  24
  25
  26
  27
  28
  29
  30
  31
  32
  33
  34
  35
  36
  37
  38
  39
  40
  41
  42
  43
  44
  45
  46
  47
  48
  49
  50
  51
  52
  53
  54
  55
  56
  57
  58
  59
  60
  61
  62
  63
  64
  65
  66
  67