Qiang.F

icon company University of Electronic Science and Technology of China icon location Chengdu icon location A student of University of Electronic Science and Technology of China. AI4Sci & CV.

Results 6 comments of Qiang.F

期待出更大尺寸的模型

> Thanks for your attentions. m/l/x will come soon. 请问出来了吗？

测试精度比YOLOv5低8个百分点

是的，在我自己的数据集上比yolov5也低了8-10个点

转onnx后如何对输出结果进行解析？

可以参考这个，他还写了教程[https://github.com/hpc203/nanodet-opncv-dnn-cpp-python](url)

u and v need to have the same number of columns

I've had the same problem, but I've solved it and it might inspire you. I resize the size of the picture during training, so in get _ iou(), get _...

EMSA

应该不是，除了Attention这个类，还有下边Block中的deforward（）中的第一行，Attention中我好像没发现有Residual connect，如果加上Block才是完整的EMSA，还有这个EMSA（v1）和CMT中的LMSHA结构非常像，那个图画得更容易理解一些， 如有不对，请指出。

hard encoding BC in the pipe flow problem

Hello, I didn't understand this place either, also when computing the loss function, it doesn't seem to follow Eq. 10. It would be appreciated if you could tell me about...