About the training data

[Janezzliu]

  你好，我前些天也用SSD算法识别安全帽，不过是基于tensorflow的，准确度大概70％，我想知道你是怎么制作训练样本的，不知道能不能发几个样本图片到我的邮箱？[email protected].顺便想问下你识别的准确度是多少啊？最后谢谢你分享到github上。

[HarminDo]

请问现在可以公开数据集么？或者可以私发我一下么？好像目前没有公开的安全帽检测的数据集，我也是学生.保证不做他用~谢谢啦

[komalpharate]

Hello, The pretrained models are having good acuracy. Can we get the training steps and training data?

[wujixiu]

Hi, everyone, the training data will available sooner or later for more data are added to the dataset.

[amenegola]

Hi there, @wujixiu !

Really nice code and trained networks. For medium to large objects, it works perfectly!

I would like to know, if possible, more details about your training data. This would help me to build my own dataset. I would like to know things like:

• How many images does your training set have?
• How did you annotate your images? By hand or with some sort of pre-detection?
• How much is your positive/negative samples ratio?

Really nice work, congrats!

[wujixiu]

Hi, @amenegola

• Our dataset provides 3,174 images containing 18,893 hardhats instances (including non-hardhats regions).

• All images are labelled mannually with labelImg.

• Same pos/neg ratio in SSD. Training Data will be released at the end of this month.

[amenegola]

Hello @wujixiu!

Awesome, thanks for this reply, you are very helpful.

Congrats again on the model!

Cheers,

Afonso

[codeqby]

I'm looking forward to the training data, too. Thanks for your great job.

[YuGuii]

非常感谢你所做的贡献，不知道博主是不是可以公开一下数据集，同时我也有些问题想要请教一下博主，我的QQ是1018949282，同时再次非常感谢博主！