Open GluttonK opened 10 months ago
can you schedule a remote meeting with me, and I'll explain the meaning of this code to you
Thank for your answer . Here is my email and QQ : 2287512349@qq.com You can send an email or Add Friends with me to schedule a remote meeting I will set aside sufficient time to wait for your reply
I also have the same confusion about the code, it seems it only work with kernel size of 3. Can you give more explaination about it? Thanks very much.
@Anm-pinellia @GluttonK Hi, this can be taken a fast implementation of torch.nn.functional.affine_grid
and torch.nn.functional.grid_sample
. These two operations can be simplified when only rotating a 3x3 kernel.
我也有点confused 这个代码,最后通过这个卷积,输出的特征图 是已经纠正旋转的特征图吗
I don't understand the rot_mat_positive and rot_mat_negative. why can we get the rot_mat by this way?maybe my math is bad.
Hi, the math derivation behind this code snippet is quiet complicated, so that I cannot make it clear by using several sentences. I will make some slides to show the derivation process when I'm free (maybe in the summer vacation).
OK, thank you very much.
For the mathematical formula here, I also have some confusion. Could you please provide a detailed derivation of the formula?
Hello, I don't quite understand the principles of obtaining the matrices rot_mat_positive and rot_mat_negative. Could you please explain? Thank you.
looking forward to further details from authors
`def _get_rotation_matrix(thetas): bs, g = thetas.shape device = thetas.device thetas = thetas.reshape(-1) # [bs, n] --> [bs x n]
在这个生成旋转矩阵的函数中,我不太理解得到rot_mat_positive 和rot_mat_negative这两个矩阵 的拼接原理。这只适用于旋转3x3标准卷积,我希望理解这段代码以试图将其扩展到更大的卷积上。 能否提供一些理论上的支持,非常感谢