support for non-centered cameras #75
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Hello,
I've added support for non-centered cameras based on the answer you gave me here.
I've tested the changes on a few examples of my dataset and it seems to work ok.
I have however a few concerns:
1/ The training time is rather long (a bit less than 4 hours) with about 2 it/s. I'm using an rtx A6000 for reference, and my dataset has 68 cameras at 2k resolution (I'm training with the -r 1 option).
2/ The memory usage keeps increasing even after 15000 iterations, despite using the default value for densify_until_iter, and it reaches about 30GB.
3/ The geometry is good for regions of the surface with uniform texture whereas parts with high frequency texture patterns tend to be reconstructed more poorly.
4/ I've managed to use the SIBR remote viewer to visualize the training without issue but the non-remote viewer crashes on the trained model (even with the mipnerf360 dataset).
Is any of this expected?
I've searched for all occurrences of "H/2" and "W/2" to replace them with cx and cy, but more changes could be necessary since I'm not familiar with the codebase. Can you review my modifications and give me some advice in case I missed something?
Thanks :)