2.5 KiB
Minutes from 54th OpenVDB TSC meeting, June 30th, 2020, (GMT)
Attendees: Nick A., Jeff L., Ken M., Dan B.
Additional Attendees: Johannes Meng (Intel), JT Nelson (Blender), Peter Cheng (DW), Andre Pradhana (DW)
Regrets: Peter C.
Agenda:
-
Confirm quorum
-
Secretary
-
New Meeting Time
-
7.1 Fast Sweeping
-
7.1 Sharpening
-
Dilation Improvements
-
Next meeting
-
Quorum was confirmed.
-
Secretary was Nick Avramoussis
-
New Meeting Time
New meeting time to be scheduled an hour earlier. Every Tuesday at 10am PST / 1pm EDT / 6pm BST. Johannes to check with Bruce about a possible conflict.
- 7.1 Fast Sweeping
Jeff and Nick have posted reviews on the PR. Jeff, various comments on input types and exception behaviour. Nick, primarily concerned about const casting and redundant code. Ken to attempt to address the feedback. Andre had tested the new SDF dilation based algorithms which are working and will continue to work with Jeff to prepare a SOP component. Noted that vector grid processing still requires per component extension via vector splits and merges. Ken to see how feasible it would be to avoid this step. Dan, CI builds are still failing on this PR, requires local builds to build with -DOPENVDB_CXX_STRICT to reproduce the problems.
- 7.1 Sharpening
Ken, approved implementation but unable to checkout/copy peters changes to ASWF/fork so still needs unit tests. Two ways to resolve this, either clone the fork or merge into a temporary branch in the ASWF repo. Dan to take a look at achieving the latter. Nick, concern that no unit tests now mean none later. Ken to take another look at the result of Dan's merge of Peter's branch. Touched again on the fact that Peter's implementation of the sharpening kernel is not separable and thus trades some performance for more robustness. Ken, Peter's benchmarks look promising and the alternative implementation (unsharpen masking i.e. signal processing, low pass filter to remove high freq) may not work on levelsets.
- Dilation Improvements
Nick, new PR which fully implements the proposed changes from #675 will be up shortly. Aims to deprecate the existing dilateVoxels and erodeVoxel methods with new functions that provide a more uniform interface and support matching functionality. Adds support for edge/vertex erosion based on inverse dilation. Significant performance improvements to large dilations. Note that the result of the threaded dilation differs due to the behaviour of topologyUnion. This is addressed by #751.
- Next meeting
July 7th, 2020. 1pm-2pm EDT (GMT-4).