CS-Jackson

I reduce the default learning rate to 1e-3. During training, it still start with 1e-2. Any suggestions?

你好，请问SubReactor具体是用什么对象来表示的呢？

3

还是用EventLoop，还是使用epoll或者Channel？

Something I want to ask.

1

During inference, I notice that you first resize the picture as input, but the annotations used in counting mAP doesn't change, is it right, or do I miss something?

I can't download the trained model from google driver. Can you share the model on baiduyun?

你好，我想请教一下CNDBpedia.spo这个三元组的制作方式？我使用您共享的文件里面的词语进行检索，会出现多个关系组，请问您是根据什么规则进行过滤的呢？还有就是使用CNDBpedia.spo的网站提供的API是否需要申请apiKey以获得大规模调用的功能？

Why did my test only achieve mAP: 0.409?

28

I tried the `python test.py --weights_path weights/yolov3.weights`, but get mAp: 0.409

How to reproduce the result in README?

2

I download the dataset produced in #5 ,and try to reproduce the result showed in README. However, the train accuracy stuck in 0. Can you give me some details about...

Can you shared some details about trained BiseNet from scratch?

1

Did you use the same setting with the pretrained Xception39?

What's the different between MAdd and the GFlops which are showed in the output? I look at the ResNet paper and found that the MAdd is equal to the GFLOPs....

In your script, the parameter 'momentum' in Batchnorm2d has been set to 0.0003(default=0.1). Do you use it during training, or just for inference?