目前有很多的开源实现，基于各种方法对YOLOv3进行裁剪和压缩，例如：
SlimYOLOv3
YOLOv3-model-pruning
YOLOv3-complete-pruning

• 因为我自己训练的yolov3模型是基于darknet的C语言版本训练的，生成的模型是.weights文件，一般模型压缩使用的都是pytorch或者tensorflow框架的。
• 于是第一步需要将.weights文件转换成pytorch模型pt文件。（相对tensorflow我更熟悉pytorch一点，所以就转换pytorch模型了。）

模型转换的项目：
convert_yolo_weights
转换脚本

if __name__ == '__main__': convert(cfg='cfg/yolov3.cfg', weights='weights/yolov3.weights')

转换后生成convert.pt
然后就可以基于这个convert.pt模型文件进行模型压缩了。
模型转换完成后需要测试一下是否转换正确，预测一下输出，使用detect.py测试，测试结果如下：

文章图片

这样看来模型转换结果没有问题。
下一步就需要考虑模型压缩的问题了。
参考：YOLOv3剪枝压缩99.04%参数，精度不变

• 测试模型剪枝(因为无法上传svg图片，只好上传cfg文件了)，我的模型只是从234M压缩到了33.7M，模型大小压缩到了原模型的14.4%。
• 主要是filter的个数减少了，网络结构没有改变。模型大小比33.1M的yolov3-tiny大一点点，效果高出不少。现在只希望把33.7M的模型继续压缩一下。
  原yolov3模型cfg

[net] # Testing # batch=1 # subdivisions=1 # Training batch=64 subdivisions=8 width=416 height=416 channels=3 momentum=0.9 decay=0.0005 angle=0 saturation = 1.5 exposure = 1.5 hue=.1learning_rate=0.001 burn_in=1000 max_batches = 500200 policy=steps steps=400000,450000 scales=.1,.1[convolutional] batch_normalize=1 filters=32 size=3 stride=1 pad=1 activation=leaky# Downsample[convolutional] batch_normalize=1 filters=64 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear# Downsample[convolutional] batch_normalize=1 filters=128 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear# Downsample[convolutional] batch_normalize=1 filters=256 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear# Downsample[convolutional] batch_normalize=1 filters=512 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=512 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=512 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=512 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=512 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=512 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=512 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=512 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=512 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear# Downsample[convolutional] batch_normalize=1 filters=1024 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=512 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=1024 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=512 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=1024 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=512 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=1024 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=512 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=1024 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear######################[convolutional] batch_normalize=1 filters=512 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=1024 activation=leaky[convolutional] batch_normalize=1 filters=512 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=1024 activation=leaky[convolutional] batch_normalize=1 filters=512 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=1024 activation=leaky[convolutional] size=1 stride=1 pad=1 filters=24 activation=linear[yolo] mask = 6,7,8 anchors = 20.11,297.45, 23.75,58.87, 26.94,140.45, 40.95,319.31, 57.66,23.88, 139.99,29.94, 148.85,15.06, 303.99,20.82, 320.36,41.11 classes=3 num=9 jitter=.3 ignore_thresh = .7 truth_thresh = 1 random=1[route] layers = -4[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[upsample] stride=2[route] layers = -1, 61[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=512 activation=leaky[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=512 activation=leaky[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=512 activation=leaky[convolutional] size=1 stride=1 pad=1 filters=24 activation=linear[yolo] mask = 3,4,5 anchors = 20.11,297.45, 23.75,58.87, 26.94,140.45, 40.95,319.31, 57.66,23.88, 139.99,29.94, 148.85,15.06, 303.99,20.82, 320.36,41.11 classes=3 num=9 jitter=.3 ignore_thresh = .7 truth_thresh = 1 random=1[route] layers = -4[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[upsample] stride=2[route] layers = -1, 36[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=256 activation=leaky[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=256 activation=leaky[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=256 activation=leaky[convolutional] size=1 stride=1 pad=1 filters=24 activation=linear[yolo] mask = 0,1,2 anchors = 20.11,297.45, 23.75,58.87, 26.94,140.45, 40.95,319.31, 57.66,23.88, 139.99,29.94, 148.85,15.06, 303.99,20.82, 320.36,41.11 classes=3 num=9 jitter=.3 ignore_thresh = .7 truth_thresh = 1 random=1

剪枝后模型yolov3-prune.cfg

[net] # Testing # batch=1 # subdivisions=1 # Training batch=64 subdivisions=8 width=416 height=416 channels=3 momentum=0.9 decay=0.0005 angle=0 saturation=1.5 exposure=1.5 hue=.1 learning_rate=0.001 burn_in=1000 max_batches=500200 policy=steps steps=400000,450000 scales=.1,.1[convolutional] batch_normalize=1 filters=10 size=3 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=11 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=16 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=11 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=14 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=33 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=14 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=29 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=14 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=19 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=49 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=19 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=19 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=59 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=19 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=55 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=19 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=50 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=19 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=60 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=19 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=55 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=19 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=66 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=19 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=48 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=119 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=48 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=117 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=48 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=126 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=48 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=123 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=48 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=133 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=48 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=130 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=48 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=130 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=48 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=131 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=48 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=166 size=3 stride=2 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=250 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=166 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=292 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=166 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=297 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=166 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=295 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 filters=166 size=3 stride=1 pad=1 activation=leaky[shortcut] from=-3 activation=linear[convolutional] batch_normalize=1 filters=238 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=620 activation=leaky[convolutional] batch_normalize=1 filters=252 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=621 activation=leaky[convolutional] batch_normalize=1 filters=245 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=699 activation=leaky[convolutional] batch_normalize=0 size=1 stride=1 pad=1 filters=24 activation=linear[yolo] mask=6,7,8 anchors=20.11,297.45, 23.75,58.87, 26.94,140.45, 40.95,319.31, 57.66,23.88, 139.99,29.94, 148.85,15.06, 303.99,20.82, 320.36,41.11 classes=3 num=9 jitter=.3 ignore_thresh=.7 truth_thresh=1 random=1[route] layers=-4[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=leaky[upsample] stride=2[route] layers=-1, 61[convolutional] batch_normalize=1 filters=100 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=260 activation=leaky[convolutional] batch_normalize=1 filters=108 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=295 activation=leaky[convolutional] batch_normalize=1 filters=144 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=426 activation=leaky[convolutional] batch_normalize=0 size=1 stride=1 pad=1 filters=24 activation=linear[yolo] mask=3,4,5 anchors=20.11,297.45, 23.75,58.87, 26.94,140.45, 40.95,319.31, 57.66,23.88, 139.99,29.94, 148.85,15.06, 303.99,20.82, 320.36,41.11 classes=3 num=9 jitter=.3 ignore_thresh=.7 truth_thresh=1 random=1[route] layers=-4[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky[upsample] stride=2[route] layers=-1, 36[convolutional] batch_normalize=1 filters=61 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=140 activation=leaky[convolutional] batch_normalize=1 filters=63 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=150 activation=leaky[convolutional] batch_normalize=1 filters=70 size=1 stride=1 pad=1 activation=leaky[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=241 activation=leaky[convolutional] batch_normalize=0 size=1 stride=1 pad=1 filters=24 activation=linear[yolo] mask=0,1,2 anchors=20.11,297.45, 23.75,58.87, 26.94,140.45, 40.95,319.31, 57.66,23.88, 139.99,29.94, 148.85,15.06, 303.99,20.82, 320.36,41.11 classes=3 num=9 jitter=.3 ignore_thresh=.7 truth_thresh=1 random=1

【yolov3模型压缩】参考：
参考：

推荐阅读

• 

  电商如何做项目管理经理，电商项目主管
• 

  阳姜怎么吃有什么功能
• 

  显示屏隔着纸箱摔倒会碎吗 玻璃饭盒会摔碎吗
• 

  2023年央视春晚周深参加吗 2021央视春晚周深
• 

  蚂蚱是怎么繁殖的
• 

  大雪养生的意思 大雪养生有“七宜”保健康
• 

  AndroidStdio小米手机安装APK失败
• 

  卧室婚纱照怎么挂好看 婚纱照怎么挂
• 

  杜比视界播放器,mac杜比视界播放器
• 

  不得不爱伴奏有歌词 不得不爱伴奏
• 

  非常难过、委屈、焦虑的一天
• 

  如何删除Word中的多余的空格
• 

  松下定速挂机空调显示e2是什么原因造成的
• 

  惊艳了我的科学训练
• 

  Re:《Unity|Re:《Unity Shader入门精要》13.3全局雾效--如何从深度纹理重构世界坐标
• 

  国内如何申请读博士生 国内如何申请读博，国内如何申请读博士
• 

  emui11|手机换机寿命榜出炉，小米垫底，华为未进前三，第一意料之中
• 

  微信红包2021封面免费领取 2023免费微信红包封面领取入口
• 

  猫为什么会突然在床上拉尿是生病了吗 猫为什么会突然在床上拉尿
• 

  苏州必打卡的景点 苏州最值得打卡的2个景点

• 深度学习|YOLO v4网络实现及解析
• 目标识别|YOLO(1)--------------NO GPU模式下配置、使用yolo v3
• 计算机视觉|手把手教学Yolo3的安装和实现
• Pytorch|使用google colab + yolov3_tiny(pytorch) 训练自己的目标检测数据集