HitNet

HitNet copied to clipboard

训练细节

1

您好，这个工作特别出色，很有启发意义，我有一点小疑问，论文中提到训练经过了100个epoch，但是我实际训练的时候，大概25个epoch后就收敛了，后面的训练loss和验证mae几乎没变化了，这是什么原因？期待您的解答，谢谢！

whyandbecause

Code

4

Hello, I am looking forward to the release of the source code for your model. Thank you very much!

Caohz678

How to convert pth model to onnx?

1

My python code for converting the model is as follows, but I don't get the onnx model ``` import torch import torch.onnx from lib.pvt import Hitnet def pth_to_onnx(input, checkpoint, onnx_path,...

meifannao

我想问一下：为什么GT使用双线性插值而不是最近邻插值调整尺寸？

1

LYKlyk

utils/dataloader 32行代码有问题

1

utils/dataloader里面的第32行代码，if self.augmentations == 'True'逻辑有问题：self.augmentations数据类型是bool类型，而'True'是字符串类型，所以条件会一直判断为错。

xiaohainku

Which of the pvt.py and pvtv2.py in the code is the network model?

dante1024

Problem in the 'multi-resolution feedback refinement'

1

Wonderful work! However, is the second $Y_1^{in}$ refers to $Y_3^{in}$ in this section? 

wangtong627

We would be very grateful if the code could be released, we are all looking forward to it.

1

ipanda555

We would be very grateful if the code could be released, we are all looking forward to it.

25252www

Nice work in AAAI23! We are looking forward to the code.

ChunmingHe