search

yinboc / liif

Learning Continuous Image Representation with Local Implicit Image Function, in CVPR 2021 (Oral)
https://yinboc.github.io/liif/
BSD 3-Clause "New" or "Revised" License
1.28k stars 145 forks source link

question about detail #21

Closed XiaoyuShi97 closed 3 years ago

XiaoyuShi97 commented 3 years ago

https://github.com/yinboc/liif/blob/68d6164f9c200e44861bb74ad489c60bbff77fbb/models/liif.py#L82 Hi, to my understanding, rel_coord here should be normalized (i.e. in range [-1,1]). Why do you multiply it with feature size? Thx!

yinboc commented 3 years ago

Hi, your understanding is correct and this is the normalization. Before this line, "rel_coord" has the range [-1/H, 1/H] and [-1/W, 1/W], where H,W is the shape of the feature map, thus we normalize by multiplication.

XiaoyuShi97 commented 3 years ago

I see. Thx for your reply.

shkarupa-alex commented 2 years ago

Hi, your understanding is correct and this is the normalization. Before this line, "rel_coord" has the range [-1/H, 1/H] and [-1/W, 1/W], where H,W is the shape of the feature map, thus we normalize by multiplication.

Found my mistake in previous version of this comment. You were right, sorry.

XiaoyuShi97 commented 2 years ago

你好，这是石晓宇的邮箱。我会尽快处理该邮件