Jialiang Ma

> 是的，作者在这里是把C4->C5的dilation修改为了2(默认为1)，但是我自己做实验的时候发现设置为1的话，效果会好一些。 > 这里是我的一个实现，不过是基于paddle的，不过暂时没有添加CxAM和CnAM模块，你可以看下 @shaohailin > https://github.com/PaddlePaddle/PaddleDetection/blob/master/configs/acfpn/README.md > https://github.com/PaddlePaddle/PaddleDetection/blob/master/ppdet/modeling/backbones/acfpn.py Hi, I have been trying to reproduce this paper using mmdetection, I refer to this repository and your version of paddlepaddle...

I checked my experiments and found that the mAP is very low when training with multiple GPUs, but the mAP is normal when only one GPU is used. May I...

> > I checked my experiments and found that the mAP is very low when training with multiple GPUs, but the mAP is normal when only one GPU is used....

> Yeah this is strange - when I go to previously working versions I still get: > > ``` > from pydantic.fields import FieldInfo, Undefined > ImportError: cannot import name...

> > > Yeah this is strange - when I go to previously working versions I still get: > > > ``` > > > from pydantic.fields import FieldInfo, Undefined...

Well, these images are obtained by _tensorboard_. And the codes to merge all the losses into the tf.summary are written in the original materport repository, you don't need to change...

I have the same question, and I translate this issue as follows: " Hi, would like to ask you a question. I don't think it's reasonable to use centroids-related logic...

Still broken...