Xiaoyu Li

I guess it is right, because (d1+d2+d3)/3 *0.3 = (d1+d2+d3) * 0.1

@Laviyy You could try to set the value of gamma to 0.0002.

@Laviyy Another tip is setting the value of 'type' to SINGLE, at the first.

Maybe the batch_norm layer could give a fixed value , when the result of the layer is Nan.

@zhangwei147258 Sorry, the project includes models about emotion.[https://github.com/jimeffry/Age-gender-and-emotion-recognition](url)

You could refer to the repo, https://github.com/jimeffry/face-detect-recognize, or this one https://github.com/jimeffry/emotion-and-gender-classification for detecting faces.

搭建那会没有找到mobilenetv2的tensorflow预训练模型，后续我没有再训练。多目标检测若没有预训练模型，很难收敛。建议转到pytorch，有好多预训练模型可用。

搭建那会没有找到mobilenetv2的tensorflow预训练模型，后续我没有再训练。多目标检测若没有预训练模型，很难收敛。建议转到pytorch，有好多预训练模型可用。

The model of tensorflow is hard to convergence,because no pretrained model. I recommend the mxnet model to you. Please read the first line in ReadMe.

It's a pity, I have not finished training. When it is finished I will uplaod.