forked from opencv/opencv
-
Notifications
You must be signed in to change notification settings - Fork 6
2018
Gary Bradski edited this page Feb 21, 2018
·
24 revisions
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- Name
- todo
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- Go over federal non-profit org
- OSCON
- Brainstorm on new focuses
- OpenCV release and 4.0
-
Org filing
-
New release
- 3.4.1 Release end of week
- Support for quantized models in DNN
- Face detection module is now 2.7M, works way better than Haar
- Many patches and fixes
- Likely last release of 3 series, work starts on 4.0 version
- Support for 3.4 will not stop, it will be supported for some years
- 4.0
- Sometime mid-July.
- Will have wiki linked to GIT feature "evolution" proposals
- That people or corporations can fund
- And will form a more GIT linked feature list for future GSoC Ideas list
- Submit feature request to GIT bug tracker request
- Will be on wiki -- all a github repository
- Ask Maxime where Wiki images should go
- HighGUI can we combine 2D and 3D, camera trajectories (uses VIS right now)
- How to render fonts ... text rendering ... use gtk or other
- Replace QT functionality, write real buttons
- Several proposals to put this together
- Will put it on the new "evolution" issues page to come.
- 3.4.1 Release end of week
-
OSCON
- Beta or Alpha 4.0 release will be out
- Tutorial -- intro future of opencv, new org, new feature request ways to join/help
- DNN -- face tracker, other (how to squash a net??)
- Python interface use
- Some examples that support ... interactive art (face, body tracking) or robotics
-
New focuses
- Robotics
- Support First Robotics with a basic recognition kit (2D, 3D ...)
- Courseware kit?
- Arts
- Interaction
- Robotics
- Gary
- Org setup
- Get paypal data to accountant
- Vadim
- Ask Maxime (spelling?) how to add images to the OpenCV wiki/website. Now that it is GIT based, you have to link to an image somewhere, can be a specialized GIT repository for this.
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- Happy Valentines Day!
- New meeting time, 9:30am Wednesday Californa time
- GSoC
- Idea list
- Org structure
-
GSoC
- Didn't get in this year ... we suppose the reasons (in order):
- Our ideas list was not compliant ... no real excuses but:
- We are an older org, so only one grad student on our board, the rest are all working full-time professionals and/or running companies etc and we just picked up on the email warning about Google being serious about this item this year too late.
- As with most of these items, one problem is that we've been in for many years and were just following patterns that had worked, but some things have changed, the emphasis on this is one, another rating point in the summer and so on.
- Corrective: is simply below -- make the list compliant
- We were late in the 2017 mid-summer student rating, again, no excuse, but we were in the pattern from the previous year where this didn't exist and several of us just happened to be traveling.
- Corrective: (1) Is to make sure this doesn't happen, (2) I'm trying to get more money into the org so that we can hire some full time staff who can keep up with the greater demands on the org.
- We've been in for years -- they might have also decided we needed a kick/or to just rotate us out
- Corrective: Just apply again next year.
- Our ideas list was not compliant ... no real excuses but:
- Didn't get in this year ... we suppose the reasons (in order):
-
Idea List
- Have to get compliant with Defining a Project Ideas list from the Mentor Guide
-
Org Structure
- Traffic is up, OpenCV 4.0 is coming out this summer, we have major new content to digest from opencv_contrib and particularly incorporation of of the deep neural network module DNN.
- It's time to take the running of the org up a level:
- Refile the federal non-profit status, include hosting standards such as
- Certifying camera/smart camera modules
- Image datasets standards
- Selling or partnering for education kits
- Recruiting or changing over to a more professional business and technical board
- Changing law firms to a bigger/scarier law firm that has offered to handle this pro-bono
- Recruiting a business manager and probably admin
- There is promised funding for some of this, need to convert promises to checks
- As a stretch goal, maybe run OpenCV Summer of Code at a small level this year:
- DNN tutorial/ONNX to work well with ingesting major deep platform networks.
- python
- highGUI
- Refile the federal non-profit status, include hosting standards such as
- OSCON
- Asked for a 3 hour tutorial ... Need to see if we can find someone to help or turn it down. Would like to say "yes"
- Open3D
- Will be meeting with vladlen koltun about Open3D on Feb 15th
- Gary
- [~] Org setup started...
- Vadim
- See if we have people to run the OSCON tutorial
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- GSoC
- Halide
- Open3D
- GSoC
- HighGUI person
- tiny-dnn policy
- Halide plans/schedule
- Open3D
- Will be meeting with vladlen koltun about Open3D on Feb 15th
- Gary
- Org setup
- [-] Ask Yann about students for ONNX task
- Vadim
- [-] Send name of contact person
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- Change meeting times
- DNN
- New meeting time, 9:30am Wednesday Californa time
- DNN
- Backend (fusion, convolution)
- C++
- SSE, Neon ...
- API is general
- OpenCL
- Halide (could generate CUDA, OpenCL or anything)
- AMD graphics
- Backend (fusion, convolution)
- DNN work
- Halide ongoing
- Gary
- Org setup
- Ask Yann about students for ONNX task
- Vadim
- Send name of contact person
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- GSoC Projects, mentors
- Now applied to GSoC
- GSoC Projects, mentors
- tiny-dnn and DNN
- How to resolve DNN and tiny-dnn
- DNN runs inference
- ONNX
- Integration with TVM ... relationship to Halide
- tiny-dnn and DNN
- Opencv 4.0
- DNN integrated into core as a central function
- Somethings move to conrib
- Geometric vision differences with deep learning
- Deep learning
- DNN runs inference
- Will stay in OpenCV
- Will not have any dependencies
- Can add extensions ... needs NVidia
- 4 backends:
- C++ +intrinsics optimizations;
- OpenCL;
- Halide;
- DL Inference engine
- Can we just open to different backends to allow extensions to NVidia, ARM?
-
TVM?
- But whole LLVM is in it
-
TVM?
- deep ideas:
- it would be nice to have self-contained ONNX importer (using libprotobuf?).
- some network compression algorithms, probably including quantization and factorization-based methods (where compression algorithm itself can be implemented as pytorch or Caffe2 or TF script, but then OpenCV probably needs to be extended to parse such compressed networks).
- we would also be interested in adding new layers that would help to support new interesting topologies. In particular, it would be cool to have supported some compact networks for text detection and for background segmentation, maybe shadow elimination, tracking etc. or whatever Yann considers to be worthy.
- Gary
- Org setup
- Pay
- Ask Yann about students for ONNX task
- Vadim
- Send Edgar OpenCV slides for U. Barcelona presentation
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- GSoC
- GSoC
- OpenCV has officially applied to be an Org for 2018!
- Optimizing nets ... can be hard
- Mobile net, u net for segmentation, for example VGG replaced by Mobile
- Replace layer by layer with quantization, pruning
- Fine tuning ... need to train
- Hardnet descriptors
- Features
- Add support for ONNX ...
- Translating ONNX structures to be run in DNN
- TensorFlow uses different memory layout than caffe
- Smaller semantic differences
- We have a good mentor for this, if we can find the student
- MXnet ?
- Generic compilation inside of graph
- Halide part to generate ops and scheduling
- Python interface Keras API, CNTK
- Possible source of imagery from ...
- Judge appropriateness or not of images
- Judge affect of images on users
- General applicability
- Fever seems over from flu...
- Gary
- Org setup
- Pay
- Ask Yann about students for ONNX task
- Vadim
- Send Edgar OpenCV slides for U. Barcelona presentation
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
- GSoC 2018
- Org ideas
-
GSoC
- Categories
- Python
- highGUI improvements
- C to C++ translator
- SLAM
- Deep model curation/optimization
- Better TensorFlow, Pytorch integration
- Categories
-
Org
- Fully funded non-profit
- Business basis
- Image analysis
- Seminars, workshops
- Education, documentation
-
Gary has the flu despite the vaccine ... and a headache.
- Gary
- Set up ideas page
- Set up 2018 Notes
- Apply for GSoC 2018
- Org setup
- Pay
-
Home
- Changelog (older)
- New functionality discussion
- Android
- CiteOpenCV
- OpenCVLogo
- Deep Learning in OpenCV
- OpenCV 3
- Development process
- Tutorials
- Computer Vision and Pattern Recognition
- Google summer of code
- Vision challenge
- Workshops