5.4 KiB
Minutes from 23rd OpenVDB TSC meeting, July 11, 2019
Attendees: Dan B., Peter C., Ken M.
Absent: Nick A., Jeff L.
Additional Attendees: Bruce Chernia (Intel), John Mertic (LF)
Agenda:
-
Quorum
-
Secretary
-
Round-Robin: Current and upcoming PRs
-
Graduation
-
Next Meeting
-
Quorum was confirmed.
-
Secretary was Dan Bailey
-
Round-Robin: Current and upcoming PRs
Ken:
Fast sweeping, value extension tool to be shared very soon. Initially this will just be the core library implementation with subsequent discussion about the right approach for a Houdini integration to follow.
Peter:
PR 455 - Some issues raised on the Sharpening PR that need consideration. Optimization to repopulate stencil in Z-axis could be generalized to more axes and extended to support jumps of more than one voxel. Also, Peter has been asked by artists at DWA for a more generic convolution kernel and this PR could be seen as a proof of concept of doing that.
Ken proposed (and has already implemented) an alternative technique which is well established (actually dates back to the 1930s) and often referred to as "unsharp masking". Peter wasn't able to get satisfactory results using this technique, however that does not mean that it doesn't work - it certainly does for Ken. However, it suggests that Peter might have found a better solution which of course remains to be proven. Both methods are based on separable filter kernels so Ken doesn't see why there should be a significant difference in the computational efficiency, but cannot speak to the difference in quality since he hasn't had a change to do a direct comparison.
The current implementation only works on level sets, but could be relatively easily extended to support fog volumes and vector fields. Ken mentioned that vector fields would be very useful to him. One option is to complete the current PR and then introduce a new PR for the extension.
Peter didn't have access to his list of forthcoming work.
Dan:
PR 487 - Fixes CI issue where Houdini builds are skipped rather than fail when run outside of ASWF Azure account. Trivial fix, anyone can approve.
PR 483 - Adds code coverage support to CMake, Nick currently reviewing. Minor issue where code coverage takes a very long time to run (2 hours+) so can only be run sporadically. A number of options to resolve this. Ken suggested running the unit tests concurrently, though there may be issues if the tests themselves run in parallel. Dan suggested using a smaller data set for some of the heavy tests, perhaps with a flag that enables the more expensive tests for a less frequent test run. Peter suggested using a mode that he added to the unit tests where you can supply a file that defines which unit tests to execute. This would allow for partitioning the tests into groups to be run concurrently across multiple vdb_test processes.
All agreed there's lots of maintenance work to do around unit tests, improving code coverage, breaking up some of the bigger tests like TestTools, etc.
PR 482 - An attempt to resolve the vulnerabilities reported by SonarCloud. Low priority PR, merely included to start conversation. This adds a log or error to empty catch statements based on whether they are expected to fail or not. Peter commented that this might be problematic for the matrix inverse in particular, Dan suggested we slightly refactor this one to not communicate with exceptions.
PR 479 - Meeting notes from last meeting, would quite like a meeting attendee to approve before merge.
PR 464 - Unification PR. Peter to look into the spare data API and how to retrieve a value previously inserted. Peter prefers exposing a non-const OP_Table than communicating the labels to hide down into the OP_Factory, Dan to change this.
PR 452 - Dan to address remaining feedback by adding syncNodeVersion support to VDB Scatter SOP to be able to add position compression option now that spare data API has been introduced.
PR 436 - Delay-loading PR. One remaining question from Peter about what happens to the metadata when loading just the metadata in the VDB file as a preliminary process to loading the rest of the file. Dan to investigate and reply to this then Peter to do a final pass before we merge.
PR 402 - Peter to look into improving VDB SOP documentation, low priority item.
Main upcoming feature includes an extension to point moving to provide merging grid functionality. Two options for exposing in Houdini, extending the VDB Combine SOP and adding a new VDB Point Merge SOP. Dan presented Nick's argument for the latter and all agreed that this is the better option considering compile times and the existing node parameterization.
- Graduation
Quick review of the graduation requirements. One remaining item needs looking at before presenting for graduation - Steve Winslow has raised an issue in the proprietary license notices included in the Houdini plugin, Dan to mail Jeff to detail the specifics. Everything else seems in good shape, so hope to achieve graduation before Siggraph.
Not necessarily a strict requirement for graduation but would be nice if all could seek to sign the digital CCLA. Dreamworks and DNeg have already signed. If the paper CLA that has been signed is not the Dreamworks-based CLA, but the more recent one, this can be imported without needing to sign a new one. Send John any signed paper CLAs to import if so.
- Next Meeting
No meeting next week, next one scheduled for just before Siggraph.
July 25th 2019. 3pm-4pm EDT (GMT-4).