Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

两类目标检测,精度为0 #12

Closed
noobliang opened this issue Dec 5, 2019 · 5 comments
Closed

两类目标检测,精度为0 #12

noobliang opened this issue Dec 5, 2019 · 5 comments

Comments

@noobliang
Copy link

noobliang commented Dec 5, 2019

单类目标检测没问题,效果很好,但是同一个数据集,我检测两个类之后,刚开始训练精度非常低,0.008,经过几个epoch就降到0了, 尝试过调小学习率,增大batch,还是这样。请问有大佬碰到过吗。
Screenshot from 2019-12-05 14-14-27

@tanluren
Copy link
Owner

tanluren commented Dec 5, 2019

你好,可以看看你训练的命令吗?有没有使用预训练权重?数据处理有没有问题,如cfg的设置和label,names等是否对应

@noobliang
Copy link
Author

你好,可以看看你训练的命令吗?有没有使用预训练权重?数据处理有没有问题,如cfg的设置和label,names等是否对应

你好, 数据集我是之前跑过yolov3-tiny的,我也查看了,没什么问题, 参数配置我也分别检查过了。不知道为什么会这样,很奇怪。
Namespace(accumulate=2, adam=False, arc='defaultpw', batch_size=16, bucket='', cache_images=False, cfg='cfg/yolov3-spp.cfg', data='data/car_crop.data', device='0', epochs=100, evolve=False, img_size=416, img_weights=False, multi_scale=False, name='', nosave=False, notest=False, prebias=False, prune=1, rect=False, resume=False, s=0.001, sr=False, transfer=False, var=None, weights='weights/yolov3-spp.weights')

@tanluren
Copy link
Owner

tanluren commented Dec 5, 2019

分类损失太高,看看label里面的id是不是有问题

@noobliang
Copy link
Author

分类损失太高,看看label里面的id是不是有问题

非常感谢作者大哥回复 ,但是我发现一个奇怪的问题。数据集我没动,关掉数据增强就没问题了,刚刚训练了几个小时,基础训练没问题了。看着损失下降很开心,哈哈。
Screenshot from 2019-12-06 11-09-57
我多次测试 ,发现只要开启数据增强,就出问题。
Screenshot from 2019-12-06 13-53-14
我的数据图片尺寸小于416, 一张图就一两个目标,而且两个类别很相似,我不知道这有没有什么影响。

@tanluren
Copy link
Owner

tanluren commented Dec 6, 2019

哈哈有意思

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants