[Feature] Support PointNet++ decode head #479
Conversation
| out_sa_xyz = [] | ||
| out_sa_features = [] | ||
| out_sa_indices = [] | ||
| out_sa_xyz = [xyz] |
There was a problem hiding this comment.
In PN++ for segmentation, we need all the sa_xyz and sa_features, so I include the first xyz here.
I have checked the codebase, only 3DSSD uses PN2MSG, and it only uses the last out_sa_xyz and out_sa_features, so adding a first element won't affect its performance.
| fp_xyz=fp_xyz, | ||
| fp_features=fp_features, | ||
| fp_indices=fp_indices, | ||
| sa_xyz=sa_xyz, |
There was a problem hiding this comment.
We need sa xyz/features in PN++ for seg.
|
I have added some comments about the changes I made above. Please check them before reviewing my code :) |
Codecov Report
@@ Coverage Diff @@
## master #479 +/- ##
==========================================
- Coverage 50.66% 50.54% -0.13%
==========================================
Files 189 194 +5
Lines 14455 14527 +72
Branches 2366 2374 +8
==========================================
+ Hits 7324 7343 +19
- Misses 6638 6691 +53
Partials 493 493
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
open-mmlab#479) * If a cuda launch error occurs, verify if cuda device requires top_k to be reduced. * Fixed lint * Clang format * Fixed lint, clang-format
Add decode_head into codebase for segmentation task. Implement a base head and a PN++ head.