search

Zzh-tju / DIoU-darknet

Distance-IoU Loss into YOLO v3
GNU General Public License v3.0
313 stars 85 forks source link

用自己的数据集训练时ciou loss下降比giou loss慢，请问怎么解决呢？ #30

Open 1343464520 opened 4 years ago

1343464520 commented 4 years ago

我的数据只有一类行人，我用的是AB版Darknet，我只调了iou_normalizer参数，从0.07改成0.15还是一样，基本没啥变化。。我是从头开始train的。求助大神！以下是我的cfg文件中参数设置： [net]

Testing

batch=1

subdivisions=1

Training

batch=64 subdivisions=16 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

learning_rate=0.01 burn_in=1000 max_batches = 50200 policy=steps

steps=40000,45000

scales=.1,.1

steps=4000,8000,12000,16000,20000

steps=900,2000,3000,5000,20000 scales=.5,.5,.5,.5,.5

cutmix=1

mosaic=1

:104x104 54:52x52 85:26x26 104:13x13 for 416

[convolutional] batch_normalize=1 filters=16 size=3 stride=1 pad=1

activation=leaky

activation=leaky

Downsample

[convolutional] batch_normalize=1 filters=32 size=3 stride=2 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -2

[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=16 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=32 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -1,-7

[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1

activation=leaky

activation=leaky

Downsample

[convolutional] batch_normalize=1 filters=64 size=3 stride=2 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -2

[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=32 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=32 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=32 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -1,-10

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

Downsample

[convolutional] batch_normalize=1 filters=128 size=3 stride=2 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -2

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -1,-28

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

Downsample

[convolutional] batch_normalize=1 filters=256 size=3 stride=2 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -2

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=128 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -1,-28

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1

activation=leaky

activation=leaky

Downsample

[convolutional] batch_normalize=1 filters=512 size=3 stride=2 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -2

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[convolutional] batch_normalize=1 filters=256 size=3 stride=1 pad=1

activation=leaky

activation=leaky

[shortcut] from=-3 activation=linear

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1

activation=leaky

activation=leaky

[route] layers = -1,-16

[convolutional] batch_normalize=1 filters=512 size=1 stride=1 pad=1

activation=leaky

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

SPP

[maxpool] stride=1 size=5

[route] layers=-2

[maxpool] stride=1 size=9

[route] layers=-4

[maxpool] stride=1 size=13

[route] layers=-1,-3,-5,-6

End SPP

[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 filters=128 size=1 stride=1 pad=1 activation=leaky

[upsample] stride=2

[route] layers = 85

[convolutional] batch_normalize=1 filters=128 size=1 stride=1 pad=1 activation=leaky

[route] layers = -1, -3

[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 filters=64 size=1 stride=1 pad=1 activation=leaky

[upsample] stride=2

[route] layers = 54

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1 activation=leaky

[route] layers = -1, -3

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1 activation=leaky

[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=128 activation=leaky

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1 activation=leaky

[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=128 activation=leaky

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1 activation=leaky

##########################

[convolutional] batch_normalize=1 size=3 stride=1 pad=1 filters=128 activation=leaky

[convolutional] size=1 stride=1 pad=1 filters=18 activation=linear

[yolo] mask = 0,1,2 anchors = 89,100, 103,136, 118,181, 132,106, 141,147, 162,194, 182,134, 211,179, 233,243 # 9488 + 2825 = 12313 classes=1 num=9 jitter=.3 ignore_thresh = .7 truth_thresh = 1 scale_x_y = 1.2 iou_thresh=0.213 cls_normalizer=1.0

iou_normalizer=0.07

iou_normalizer=0.15 iou_loss=ciou nms_kind=greedynms beta_nms=0.6 max_delta=5

[route] layers = -4

[convolutional] batch_normalize=1 size=3 stride=2 pad=1 filters=128 activation=leaky

[route] layers = -1, -16

[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=18 activation=linear

[yolo] mask = 3,4,5 anchors = 89,100, 103,136, 118,181, 132,106, 141,147, 162,194, 182,134, 211,179, 233,243 # 9488 + 2825 = 12313 classes=1 num=9 jitter=.3 ignore_thresh = .7 truth_thresh = 1 scale_x_y = 1.1 iou_thresh=0.213 cls_normalizer=1.0

iou_normalizer=0.07

iou_normalizer=0.15 iou_loss=ciou nms_kind=greedynms beta_nms=0.6 max_delta=5

[route] layers = -4

[convolutional] batch_normalize=1 size=3 stride=2 pad=1 filters=256 activation=leaky

[route] layers = -1, -37

[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=18 activation=linear

[yolo] mask = 6,7,8 anchors = 89,100, 103,136, 118,181, 132,106, 141,147, 162,194, 182,134, 211,179, 233,243 # 9488 + 2825 = 12313 classes=1 num=9 jitter=.3 ignore_thresh = .7 truth_thresh = 1 random=1 scale_x_y = 1.05 iou_thresh=0.213 cls_normalizer=1.0

iou_normalizer=0.07

iou_normalizer=0.15 iou_loss=ciou nms_kind=greedynms beta_nms=0.6 max_delta=5 chart_nano_tr12313 chart_nano_ciou

Zzh-tju commented 4 years ago

具体的工程需要你自己来调试完成，我只能给一些建议。

  1. 尝试更多的系数，iou_normalizer∈[0.07,0.5]
  2. 观察数据集的物体多为中大型还是小型，小型建议使用DIoU
  3. 控制变量的情况下，如果D/CIoU均比GIoU差，那毫无疑问使用GIoU，说明D/CIoU对该数据集失效。当然也可以看看IoU loss是何情况。我们无法保证所提出的trick在任何情况下都最优，事实上一些经典的trick如focal loss，Soft-NMS也都存在失效的情况
1343464520 commented 4 years ago

非常感谢作者的中肯建议！通过控制变量法去尝试后可能是我的数据更适合giou. 曾经也尝试过focal loss也是几乎没有提升，回头换别的任务数据集再试试。再次感谢作者！

djwilv commented 4 years ago

你画loss曲线代码能发一下吗？

1343464520 commented 4 years ago

可以直接画的，我用的AlexyAB版darknet。------------------ 原始邮件 ------------------ 发件人: "djwilv"<notifications@github.com> 发送时间: 2020年6月17日(星期三) 晚上7:54 收件人: "Zzh-tju/DIoU-darknet"<DIoU-darknet@noreply.github.com>; 抄送: "1343464520"<1343464520@qq.com>;"Author"<author@noreply.github.com>; 主题: Re: [Zzh-tju/DIoU-darknet] 用自己的数据集训练时ciou loss下降比giou loss慢，请问怎么解决呢？ (#30)