-
Notifications
You must be signed in to change notification settings - Fork 1.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Enhance] Support RGB images on ScanNet for multi-view detector #696
Conversation
Codecov Report
@@ Coverage Diff @@
## master #696 +/- ##
==========================================
+ Coverage 48.93% 48.96% +0.02%
==========================================
Files 208 208
Lines 15861 15880 +19
Branches 2543 2538 -5
==========================================
+ Hits 7761 7775 +14
+ Misses 7606 7603 -3
- Partials 494 502 +8
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We would better add some basic comments or docstrings to extract_posed_images.py
and add the corresponding contents/explanations in the ScanNet dataset doc.
It is fine to borrow some official code from ScanNet with the explicit acknowledgement/citation to their credit. Just need some basic comments and explanations in the doc to make this new feature clearer. |
Hi @Tai-Wang , Have a question about Also a bit curious about your
Sorry for not very smart questions. |
Hi @filaPro ,
As for |
Hi @Tai-Wang , |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Almost ready to be merged
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@filaPro Good job! From the ImVoxelNet paper, it seems to perform training on training set and report results on val set. So all the gt boxes it uses is axis-aligned, and test set is not involved right? If that's true, this PR is done.
I just want to make sure this point because we perform point cloud alignment in the data loading pipeline. Gt boxes in both aligned and unaligned coordinates are provided.
Sure, in ImVoxelNet paper only train and val scenes are used. Btw, not quite understand how train/val/test splits are connected with point cloud alignment? We can not process this alignment on test step if necessary? |
Good question. This is mainly because the axis-aligned matrix required to perform alignment is consider as part of the annotation by ScanNet team. As a result, test scans don't have this matrix and can't be aligned. This doc provide more details about our design of ScanNet pre-processing scripts (which I believe is a bit complicated and confusing at the first glance). BTW it reminds me that, it would be great if you can add this PR's change in the compatibility doc, so that users will know they may need to re-generate ScanNet data for your ImVoxelNet models (and maybe more multi-view method in the future :) |
Can I add a |
Yes, sure! Just follow the pattern :) Edit: I think you can refer to this part, which is very similar to your changes here. |
Ready to merge after fixing that typo. |
This PR aims to support RGB images on ScanNet for multi-view detector. More discussion in #620.
It is a bit WIP now as I also plan to add images to
ScanNetDataset
. For now I just adaptedSensReader
from original ScanNet repo. Can you please check if we are ok with it? cc @Tai-Wang