VITA-Group / GNT

[ICLR 2023] "Is Attention All NeRF Needs?" by Mukund Varma T*, Peihao Wang* , Xuxi Chen, Tianlong Chen, Subhashini Venugopalan, Zhangyang Wang

Home Page:https://vita-group.github.io/GNT

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

How to visualization the view attention?Just like Figure 6 in your paper

gjgjgjfff opened this issue · comments

Thank you for your wonderful work and look forward to your reply!

To visualize view attention, I use the attention matrix (Nrays x Npts x Nviews x Dim) from here and average across the channel dimension(Nrays x Npts x Nviews), further followed by computing the view number with maximum attention value (Nrays x Npts). Finally, the mode across points for each ray is computed and visualized (Nrays).

Unfortunately I cannot seem to find the code for it, but the above is the general idea. Let me know if you run into any problems