-
-
Notifications
You must be signed in to change notification settings - Fork 943
2019
Edgar Riba edited this page Mar 16, 2020
·
8 revisions
- Discussion with @asmorkalov and @AnnaPetrovicheva from Xperience.ai about CI infrastructure.
-
Kornia uses Circle-CI as CI system right now.
- There is one testing host and it's included into Cicle-CI plan.
- The team wants to extend testing and include GPU.
-
Releases are done by Edgar semi-automatically.
- AI Edgar: send release instructions to Alexander for review.
-
Amazon provides some AWS creadits for the project.
- The team wants to do inital testing to understand price and available resources.
- Alexander proposed test GPU workload on nightly basis to reduce cost, if it's required.
- Pre-commit testing can be CPU only.
-
Windows testing is painful.
- PyTorch team runs dedicated Jenkins-CI instance and Windows hosts with GPU.
- Windows support is not priority for the team.
-
Docker is recommended solution for Linux as build and test environment.
- NVIDIA provides NVIDIA Docker2 project to run containers with GPU support.
-
Circle-CI and AWS related help and recommendations are welcome.
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- Organization
- Community
- Infrastructure
- Roadmap
-
Organization
- Kornia accepted to be part of the PyTorch ecosystem.
- Soon officially part of the OSVF allowing us to receive funding.
- Applied to AWS open source program to get free credits
- https://aws.amazon.com/blogs/opensource/aws-promotional-credits-open-source-projects/
- Camilo will bootstrap process and start collaboration with RoboMake team.
-
Community
- Number of PRs and contributors have increased a lot.
- Make sure we have clear contribution protocols, tests. Revisit contributions docs.
- Dmytro and Edgar will work on promotional slides (after cvpr deadline).
- Benchmarks wrt to some popular functions
- not a high priority now, just to have a brief idea. Include to slides.
- Kaggle
- Dmytro contact Jeremy to talk about kaggle kernels.
-
Infrastructure
- Next release v0.1.5 December 1st.
- release a new version every two months approx.
- not broke compatibility, create guidelines.
- Discuss with Camilo/RoboMake team from AWS about using CircleCI or CodeBuild for the GPU testing.
- Anna will contact Alex to help with the GPU testing framework.
- Conda packages:
- not a priority now since we only depend on PyTorch.
- Next release v0.1.5 December 1st.
-
Roadmap
- See issue: https://github.com/kornia/kornia/issues/312
- Data augmentation
- we will go for a kornia.augmentations focusing on parametrized/differentiable augmentations.
- return transforms as Luis and Diego proposed.
- Collaboration with albumentations
- GPU testing
- prepare testing framework
- setup CI
- Epipolar geometry
- Match opencv colorspaces conversions
- Differenciable drawing
- 3D reprojections/rendering
- ping Vassilis from Scape
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- Roadmap
- Organization
- Infrastructure
- Community
- Roadmap
- GOAL: grow as community until Pytorch devcon (October 10th).
- Push code used in the paper and do pre-release (v0.1.4).
- Focus on improving documentation, examples and tutorials.
- use Colab for examples and tutorials. Contact @priba, he did some examples already.
- No new features in the short term. Plan new features roadmap after Pytorch devcon (v0.2).
- Organization
- @edgar @dmytro role as maintainers.
- discussed with @gary about joining the OSVF.
- @edgar will donate his gsoc part to kornia.
- meetings schedule: monthly.
- Infrastructure
- @chadell will help with the devops.
- mail server with google domains; @edgar owns kornia.org
- Move to CircleCI; first CPU, next GPU.
- Find CPU server for Discourse.
- Find GPU server for testing; @gary has contacts with lambda; will talk to Ethan too.
- Community
- Discussion channel:
- setup Discourse
- CPU server 24h (need to look for options)
- Website:
- Should be ready before Pytorch devcon.
- @carlosb1 proposeed Jekyll template in jekyll_kornia
- will be hosted in github pages.
- Promotion:
- need to create slides
- submit paper to arxiv
- potential places to promote:
- Pytorch devcon
- @dmytro talk at his uni
- OSVF
- Pytorch ecosystem
- Discussion channel:
- Edgar
- contact @priba @diego to help out with examples
- contact facebook to ask about resources for ecosystem projects.
- ping the c++ developer and ask about status.
- Gary
- give @edgar contacts to access to gpu servers.
- Luis
- talk with Henry from Optima sports about the servers.
The Kornia.org team