yoloonme / ema-attention-module Goto Github PK

Hi ,
I hope this message finds you well.I noticed that you have not released the code for the project yet. I was wondering if you have any plans to make the code available in the future?
Thank you for your attention, and I look forward to your response.

Best regards

为什么我复现代码发现浮点运算大小为1.62GFLOPs

EMA中包含了adaptavgpool2d的操作，这个操作导致无法将其转换成onnx模型，请问如何解决这个问题？
非常感谢！！

使用mmpretrain在resnet50中验证，并没有达到论文中的准确度，尤其是在top5准确度上，多次实验并没有超过95%

您好，看到您提供的resnet50上训练的权重文件，在一开头使用的是3×3大小的卷积核，而并非官方提供的7乘7大小的卷积核，而且3×3的卷积核只有一层，请问这样不会导致感受野不同而对最终的性能产生影响吗？

What is the complete process inference formula for EMA attention mechanism？

Hi, by inserting the attention module into backbone, will the original yolov5 pre-trained weights still be available?
I think that if the backbone structure is modified, the corresponding parameters will also change, and the weights that have been learned can not be used, and we have to start training from scratch. I don't know if my understanding is correct, and I ask the author for an answer.

What is the position of the attention module added in the network when you conduct the experiment?

论文中的模型3x3分支只使用了一个3x3卷积核，为什么文章中说能够capturing multi-scale feature representation呢？不是应该堆叠多个3x3的卷积核才能获得多尺度的特征表示吗，这点不太理解。

您好，请问有预训练代码嘛？