2025 OpenCV Technical Committee

Meeting_notes

Template

2025-mm-dd

Minutes

2025-06-04

Yuantao (OpenCV China):
- not too many updates.
- GSoC student started working on tokenizers support, but no any updates so far. In general, communications are not as active or frequent as desired.
- Researching and benchmarking various FFT implementations. So far, OpenCV appears to quite efficient on RISC-V (surpisingly, it's faster than FFTW), but PFFFT (pretty fast FFT) is even faster. But PFFFT supports 1D transforms only.
- One of the student is working on optimized for RISC-V FAST implementation. Found one problem that will be fixed soon (incorrect number of detected points).
- There is another PR is about Sobel, Canny optimization. Reviewed the patch; the students continue to improve the code.
Phil (OpenCV.org):
- CVPR 2025 next week.
  - Phil is going there. Exhibition (where OpenCV has a booth) is on 13-15 June. Preparing for it.
  - We still need a demo.
  - Shiqi is not going to attend, unfortunately.
- Security at opencv.org mailing list has been fixed. Github-sponsored activities on improving OpenCV security is on track.
- Got some openmv cameras, played with them a bit, look cool.
- ...
- Bin-picking challenge: shortlist of finalists has been formed. Final testing is going on.
Gursimar, Abhishek (BigVision):
- GSoC (computational photography project): student submitted initial PR, and we reviewed it briefly.
- PR was merged to opencv_extra, opencv, opencv_zoo (deblurring models).
- working on some acceleration projects for Qualcomm chips.
  - there are some limitations on the platforms supported (certain versions of Ubuntu, glibc).
- OpenCV_zoo: github=>HF transition is going on, but there are some glitches.
  - bandwith is not enough.
  - possible solution: transfer repository itself, without LFS items. Then move models manually. For some reason, it did not work either.
- HDR/SDR patch is still in progress. Fixed some divisions by 0. Also found issues in tonemapping functions. One picture in OpenCV tutorial cannot be reproduced using OpenCV. Localized and fixed the regression: #27392, #27396.
- Abhishek is now working on spatial AI things.
Alexander S (OpenCV.ai):
- experience some problems with merging 4.x to 5.x because of conditional compilation and refactored DNN.
- reviewed the latest patches from KleidiCV guys. Found a few bugs and some of them have been already fixed.
  - going to enable KleidiCV for Linux too, not just for Android.
- remap implementation in 4.x causes problems with HAL because of the default implementation uses lookup tables for bilinear interpolation. Fixed a few tests to take into account various HAL implementations, especially those that don't use lookup tables.
- libcamera GSoC project goes well, the first patch will likely be merged more or less soon.
  - the libcamera license (LGPL) implies to shape libcamera support as a plugin
- multi-camera calibration GSoC project also goes smoothly.
  - currently clarifying requirements on how to write proper tests for different calibration algorithms, different types of cameras or systems of cameras.
  - also evaluating quality of publicly available datasets.
- there is also good progress with CI improvements.
  - support for java and javascript documentation generation via github actions.
  - opencv-python build configurations improved to use OpenBLAS on various platforms.
- there is a growing number of requests to OpenCV about Windows support on ARM64.
  - not in the nearest release, but for the next release need to add the corresponding configuration to CI
  - need to build opencv_ffmpeg*.dll for Windows ARM64.
  - need to explore whether Windows supports something like "fat binaries" which include both x64 and ARM64 versions in a single binary.
  - question to Phil: maybe someone can sponsor it?

2025-05-28

Yuantao (OpenCV China):
- Submitted PR to fix warnings in GCC 14.x. The same PR fixes possible out-of-range access in function lookup tables.
- Improved performance of div function on RISC-V by up to 2x. PR has been submitted.
- There are PRs to opencv_zoo to contribute efficientSAM model and inpainting model. They are merged. Inpainting model references HF OpenCV site (https://huggingface.co/opencv). Maybe should archive opencv/opencv_zoo at github.
  - there is a question how to download models from github if we want to load them using opencv api. Need a version of download_models.py that would use HF.
Gursimar (BigVision):
- There is now OpenCV at huggingface: https://huggingface.co/opencv
- GSoC student started some work on chromatic aberration calibration.
Abhishek (BigVision):
- Inpainting model submitted.
- HDR PR is still in progress, there is some CI problem on Ubuntu.
Shiqi (OpenCV China):
- GSoC student is working on the new version of QR code detector, trained a new model. Quality is almost comparable to wechat. Model can be pruned and yet the quality will be the same or better. Superresolution model will also be provided to improve quality of the detected QR code.
Phil:
- Attended Embedded vision summit last week, OpenCV has been presented there (booth). It was also quite successful.
  - quite a few people ask about OpenCV on ARM, in particular on Snapdragons.
- OpenCV org applied to the github program to improve security of open-source projects: https://resources.github.com/github-secure-open-source-fund/
- Thinking of creating and promoting a containerized version of OpenCV (e.g. for AWS: https://aws.amazon.com)
- CVPR is coming up: Jun 11th-15th.
  - At least Phil, Shiqi will attend.
  - Relatively small booth, but at least we will be there.
- Bin picking challenge is almost done. Now finishing evaluation. Winners should be announced within a few days.
- Open hardware summit is happening this week: https://2025.oshwa.org/. OpenCV will not be there this year, but thinking of attending it next year.
- Fun fact - there has been robot fight competition recently: https://www.core77.com/posts/137083/China-Hosts-Worlds-First-Humanoid-Robot-Fighting-Competition. Wonder whether some of those robots use OpenCV.
Alexander S (OpenCV.ai):
- Working on HAL:
  - on some things discussed with KleidiCV team.
  - continue refactoring IPP acceleration in OpenCV, moving more stuff to HAL.
  - on some FastCV-related stuff, including proper build on Android.
- Prepared 4.x=>5.x merge, have some difficulties with basic core stuff, since there a lot of code has been rewritten to support new types etc.
- Improving Python bindings (including type annotations).
- docs generation in CI via github actions: worked on adding Java docs generation
- need to have consistent BLAS support, so we are adding OpenBLAS to Windows build.
- want to make headless version of OpenCV Python by refactoring Media Foundation support.

2025-05-21

Vadim:
- Conducted initial/welcome meetings with 3 GSoC students (together with the respective mentors).
Yuantao (OpenCV China):
- hal/riscv-rvv: refactor build process https://github.com/opencv/opencv/pull/27301
- hal/riscv-rvv: use cv_elem_size1 https://github.com/opencv/opencv/pull/27315
  - TBD: fix potential risk of out-of-boundary.
- hal/riscv-rvv: add hal interface for calcHist and implement in riscv-rvv https://github.com/opencv/opencv/pull/27332
- imgproc: vectorize mode CHISQR and CHISQR_ALT in compareHist https://github.com/opencv/opencv/pull/27334
- build: fix warnings from recent gcc versions https://github.com/opencv/opencv/pull/27337
  - There is a follow-up PR (#27343) fixing more.
- bandwith of opencv_zoo is already exhausted. Even some AI chatbots download models from opencv_zoo. Thinking of moving opencv zoo to HF.
  - Gursimar: can help with it. Apparently, HF offers a convenient tool to migrate github repo to HF.
Shiqi (OpenCV China):
- GSoC:
  - will mentor one GSoC student who will work on QR detector
  - another student contacted his supervisor, author of Aruco. They scheduled regular meetings.
  - 3rd student will be working on libcamera support in OpenCV (some work has been already been done)
- trying to get more funding, including Chinese Academy of Sciences.
Gursimar (bigvision):
- testing qualcomm's PR on FastCV.
- Abhishek: fixing unstable behaviour in HDR code.
- mcc Macbeth detector bugfix (division by zero) from Vincent has been reviewed and merged
- had a call with GSoC student on improving image quality project. Decided on the plans.
Vincent:
- tried 4.12 internally, works well most of the time.
- had 2 issues, want VP to look at:
  - https://github.com/opencv/opencv/issues/27319
  - https://github.com/opencv/opencv/pull/26629
- going to submit some more issues.
- meeting with GSoC student on image codecs will be scheduled in May (before June 2nd).
Phil (opencv.org):
- Last week was "Display week" conference: https://www.displayweek.org. The event went pretty well.
  - Many people attended, lot's of stuff was given out.
  - Next year "Display week" will be in LA, going to participate there as well.
- Have the next OpenCV webinar on Thursday on building OpenCV/vision code on Snapdragon.
- Preparing for CVPR 2025
- OpenCV is finalist of github competition for open-source projects to improve security. Decision will be made soon.
Gary:
- met with "Robot tractor" people recently. Discussing possible collaboration. Maybe it will be about preparing some courseware.
- going to continue conversation with OpenMV folks.
Alexander S (opencv.ai):
- Done:
  - Alexander Smorkalov:
    - Updated NDK for OpenCV4Android builds to version 27c in CI #246. Disabled IPP x32 for it due to compatibility issues #27317.
    - Finished FastCV update #27340, opencv_3rdparty #95, opencv_contrib #3931
    - Contacted two GSoC students on multi-camera calibration and libcamera projects:
      - libcamera is in good shape, preliminary PR has been submitted already
      - mutli-camera calibration project is going to be mostly engineering one, less related to algorithms — just polish what we already have, improve testing.
    - Reviewed, tested, helped to resolve issues: ## 27344, 27341, 27337, 27334, 27332, 27331, 27327, 27326, 27325, 27338, 27321, 27322, 27315, 27314, 27309, 27307, 27305, 27302, 27301, 27343
  - Vadim Levin:
    - Added conditional inclusion support to header parser #27325. Now we can generate Python/JS/... bindings only for a subset modules, according to CMake configuration.
  - Ivan Avdeev:
    - Works on documentation build transition to Github actions.
    - Added to section https://docs.opencv.org/test/ , also cron/script download artifacts ( prepare javadocs include )
    - Found MS Media Foundation dependency in G-API for #771. Alexander works on a solution.
- OpenCV on microcontrollers:
  - 3-step plan. Need to discuss. Would be nice to have some commercial contract to be able to invest more resources into it.

2025-05-14

Vadim:
- all 9 OpenCV GSoC projects are confirmed; started communicating with students.
Yuantao (OpenCV China):
- Loongson/build: added v_gt and v_lt with lsx and lasx to fix build: PR #27266
- Merged: Refactored hal/riscv-rvv to dramatically reduce build time: PR #27301
- Still WIP: imgproc: fixing perf regressions of c3 kernels. Will try to finalize the PR in the following week. Vadim mentioned that we may need universal intrinsics for interpolation, e.g. cv::bilinear_interpolate(const v_uint8& r, const v_uint8& g, const v_uint8& b) and something like that.
- WIP: hal/riscv-rvv: added hal interface for calcHist; implemented it in riscv-rvv for type CV_16UC1: 3X speedup on K1 and 2X speedup compared to RK3568.
Gursimar:
- color correction: CCM PR is all-green now.
- fixed unstable CRF in HDR pipeline: PR #27270 (fixed issue PR #26613)
- Abhishek submitted another PR to fix bilateral filter behaviour when color sigma/space sigma (PR #27309).
- it was some activity to fix Bayer=>RGB demosaicing using VNG, but there is no ideal solution yet: #27256
Phil:
- short update: at the conference right now; last week had the first OpenCV conference (https://www.displayweek.org), which went well.
Gary:
- already communicating with GSoC student (SLAM project) on the scope and tactics (e.g. do good reference implementation in Python instead of much slower option (in terms of time needed) with C++)
- need to contact another student on LLM.
- did workshop at SID conference (https://www.displayweek.org) — went well.
- thinking of connecting OpenCV and OpenMV (https://openmv.io) and contacted some of OpenMV folks.
Alexander Smorkolov (opencv.ai):
- CI: Ivan has prepared documentation build with github actions (previously, we used custom server for it).
- Continue working on 4.12 release.
- Working on fixing several issues.
- Done with another 4.x=>5.x merge.
- Waiting for Vadim for comments in a few PRs.

2025-04-23

Vadim:
- selected hot list of 7-9 GSoC proposals; will provide the final ratings this week.
Shiqi:
- just got back from travel, no updates this week.
Alexander Smorkolov (opencv.ai):
- CI: regular Python will be tested on macos in CI.
- CI: will repair occasionally failing downloads done by CI (IPP, ...)
- working on improving HAL:
  - fixed GEMM tests to allow accelerated versions in FastCV
  - RISC-V: all patches from Yuantao and community members have been merged.
  - moving all HAL's from opencv/3rdparty to opencv/hal.
  - probably will move some basic definitions and low-level API to opencv/hal as well (probably, including cv::parallel_for_).
- reviewed, tested, helped to resolve issues with #27249, 27246, 27245, 27244, 27239, 27234, 27229, 27228, 27221, 27201, 27230, 27238, 27184.
- going to revise Java bindings, in particular use more modern API for destructors.
Yuantao (OpenCV China):
- several PRs (RISC-V RVV HAL optimizations) have been merged.
- preparing special PR for norm in 5.x (since the code is quite different in 4.x and 5.x)
Vincent:
- no public updates this time
Abhishek:
- dependency of DNN in objdetect is optional again. In this case only traditional algorithm for color pattern can be used.
- CCM patch has been refactored. CI is now green.
Gary:
- no updates, preparing for another trip.
Phil:
- OpenCV at SID conference: preparations are going on.
  - put the related announcement at hackster.io, should be published shortly.
- now have membership at Youtube. Could promote more things at OpenCV live show's.
- bin picking challenge: https://bpc.opencv.org. Decided by extend phase 1 submissions by 1 week.
Satya:
- working with existing OpenCV sponsors to continue, maybe expand partnership. Also talking to some new folks.
- have plans for CVPR and after to try to extend OpenCV sponsorship programs, community etc.

2025-04-16

Vadim:
- reviewed ~150 GSoC proposals (only those who passed preliminary filtering), created hotlist of ~15-20 proposals
Alexander Smorkolov (opencv.ai):
- Done:
  - Alexander Smorkalov:
    - Continued IPP migration to HAL ## 27194, 27213
    - Manually tested new FastCV integration #27214 on Android and Linux. Prepared CMake infrastructure update #27234.
    - Reviewed, tested, helped to resolve issues ## 27216, 27230, 27228, 27227, 27226, 27217, 27184, 27221, 27220, 27196, 27051, contrib ## 3922, 3925, 3924, 3923
  - Ivan Avdeev:
    - Tests IPP check for Android release builds #115.
    - Tests native MacOS python in CI instead of Home Brew #233.
Yuantao (OpenCV China):
- worked a few PRs with RVV optimization: dotproduct (submitted), compare (submitted), transpose (still work in progress).
- will get back to norm for RVV HAL this week.
Gursimar (bigvisionllc):
- pushed a few changes to CCM patch. Accuracy test that checks both colorchecker detector and CCM fails in CI, but it runs fine locally.
Gary:
- attended space technology conference last week
Phil:
- the next conference (https://www.displayweek.org/event-enhancements/oscca/) where OpenCV takes part is on May 20th, preparing for it.
- bin-picking challenge phase 1 ends this week; then some extra assignments will be given (at phase 2), then there will be the final evaluation.
- working on getting storage for OpenCV stuff for OpenCV online store.
- talking to existing sponsors to renew their membership.
- OpenCV channel on Youtube is now fully setup to take donations.
- RunPod (https://www.runpod.io/) wants to add OpenCV support (e.g. model zoo, benchmarks etc.)
Shiqi:
- reviewed all ~400 GSoC proposals. Most of the proposals (60-70%, if not more) seem to be AI-generated :(
- travelling now

2025-04-09

Alexander Smorkolov (opencv.ai):
- talking to Intel IPP team. Refactoring current IPP calls in OpenCV to use official cross-vendor HAL API instead.
  - refactored some IPP calls in core module
  - suggest to delegate imgproc refactoring to IPP team
- working with Qualcomm (FastCV) to update the corresponding HAL.
- closed some automatically (AI-)generated PRs, looks like pre-GSoC noise.
- looking at libcamera patch.
- some further progress with CI improvements: 1) added opencv.js testing (building and running some regression tests, for now we have just ~10 tests), 2) refactoring mac builder config to use standard Python distribution instead of brew.
Shiqi Yu (OpenCV China):
- we've got 385(!!!) GSoC proposals; maybe some of them are ChatGPT-generated.
- students are working on QR detector.
- the progress on RISC-V is good
- on libcamera: still need to contact the author of the patch from libcamera team.
Vincent:
- no updates
Gursimar (opencv.ai):
- continue work on color correction (CCM PR). Most of the comments/issues have been resolved.
  - sample was updated, but the dnn version does not always work correctly.
  - joint test (color char detection+color calibration) has been also added.
Yuantao:
- 2 PRs have been merged this week: recip & copymask in HAL using universal intrisics and native intrinsics.
- dotproduct in HAL RVV: depending on block size, tests pass or fail.
- Considering object tracking KCF, CSRT from opencv_contrib. Deep learning algos seem to work better.
Phil:
- talking to some more potential OpenCV.org sponsors.
- in bin-picking challenge we've got more submissions.
- continue working on preparing for conferences.
- interesting information: NVidia is working on CUDA Python: https://developer.nvidia.com/cuda-python; https://thenewstack.io/nvidia-finally-adds-native-python-support-to-cuda/.

2025-03-26

Alexander Smorkolov (opencv.ai):
- all RISC-V patches submitted so far (not in draft form) have been merged.
- started to migrate current IPP acceleration from opencv/modules into dedicated hal library (opencv/hal).
- looked at the recently submitted libcamera backend for videoio #27155 (submitted by one of the authors of libcamera)
- talked to Abduraghim, he cannot participate as student in this gsoc, so we need another person on deep learning feature detection & matching.
- Ivan continues working on improving CI, e.g. opencv.js build, mac build with default python etc. He found that connection to github from within one of build servers in SusTech is unstable.
Vincent:
- fixing fuzzer issues in avif and other image codecs.
Yuantao Feng (OpenCV China):
- 2 PRs with norm diff acceleration have been merged.
- PR with copyTo with mask has been submitted. Acceleration factor is not always good. Tried special instructions on RISC-V to write data with mask, results are very promising. Will try to expand this approach.
- Thinking of optimizing StereoSGBM for RISC-V.
Gursimar:
- mcc detection PR. waiting for the final approval.
- color correction - PR is updated, done some name cleanup and API cleanup.
Shiqi:
- Orbbec continues collaboration with opencv.org.
- students continue working on qr code, aruco, libcamera. On libcamera we may have conflicts with the activity by libcamera team (see updates from Alexander above).
- going to attend CVPR.
Gary Bradski:
- we are good to go with GSoC. Now application period starts.
- won't be able to attend CVPR this year.
Phil:
- was a judge at this hackathon this weekend https://youtu.be/G5uG1W48XKE
- OpenCV will participate in https://www.displayweek.org/event-enhancements/oscca/ in May.
- preparations for CVPR are finished.
- bin picking challenge is still ongoing.
- talking to potential sponsors.

2025-03-19

Alexander Smorkolov (opencv.ai):
- Done:
  - Alexander Smorkalov:
    - Prepared 4.x->5.x merge #27045.
    - Reviewed, tested, helped to resolve issues with ##27050, 27013, 26842, 27042, 26934, 27026, 27013, 26683, 26968, 27040, 27015, 27046, 27043, 26868, 27010, 27031, 27033, 27036, 27037, 27039, 27001, 26976, 26968
  - Ivan Avdeev:
    - Formated buidl_sdk.py (Android) output for Github actions parser #27027
    - Works on Linux builds issues (ccache, connection drops, etc)
Yuantao Feng (OpenCV China):
- fixed 2 issues with norm (RVV branch and general branch).
- Q from Alexander: what are the options to update SpaceMIT software (newer Ubuntu, fresh compilers etc.)? A: will ask SpaceMIT developers
Shiqi Yu (OpenCV China):
- there are some news about upcoming server-grade RISC-V (32 cores).
- students working on QR/Aruco code detection. They are trying to train one model for all use-cases.
- another group of students work on adding support for libcamera into OpenCV (e.g. to support Raspberry Pi camera).
Vincent:
- several fuzzer issues have been fixed. PRs will be submitted shortly.
Gursimar:
- PR with MCC detector (#26906) has been updated and is ready for the final review.
- PR with color calibration is being prepared.
- Need to have unit tests in Python to test the whole color calibration pipeline.
Phil:
- still working on SID conference, tickets are already available.
- the new episode of OpenCV-related interview series has been published at opencv.org: https://youtube.com/watch?v=kH1dwktZI1I. There is also Kickstarter campaign: https://www.kickstarter.com/projects/openmv/openmv-n6-and-ae3-low-power-python-programmable-ai-cameras.
- working on getting a storage for OpenCV stuff to give out at conferences/sell.
- working on OpenCV conference at San Jose.

2025-03-05

Alexander Smorkolov (opencv.ai):
- Received multiple PRs related to RISC-V RVV optimizations, but many of them stuck waiting for Yuantao review. It's suggested to merge them now and if there are some performance regressions, follow-up PRs can be submitted to disable certain entries.
- new member of OpenCV team, Ivan, started working on improving OpenCV CI. Need to setup meeting with Yuantao, get permissions and discuss rules.
- FastCV is adoped for Linux ARM builds of OpenCV. Everything works well. Looks like any ARM CPU is supported, but Qualcomm CPUs should probably provide even better performance, so it would be nice to get some Qualcomm-equipped SoCs in our CI.
- looking at several PRs related to image codecs, patches from Suleyman in particular.
- preparing another PR to merge 4.x=>5.x.
Yuantao (OpenCV China):
- submitted PR with get_norm HAL entry and provided implementation for RISC-V RVV; moved native norm optimization from OpenCV to RISC-V HAL.
Shiqi (OpenCV China):
- there is idea to create deep detector for all kinds of codes.
- working on improving OpenCV support for RPi (mostly camera-related).
Vincent:
- Q: now many PRs require approval to run through CI. What's the reason? A: the main reason is security, contributors who are not a part of OpenCV organization, need to wait for approval to test their PRs.
Gursimar:
- created PR to opencv_contrib to separate color chart detection and color correction. The next PR will remove color chart detector from contrib. The plan is to finally migrate MCC detector to the main repository.
  - it's suggested to just comment off broken/duplicate stuff in opencv_contrib/mcc to make CI green, because all this stuff will be moved to the main repository anyway.
  - look at https://github.com/opencv/opencv_contrib/issues/3316.
Phil (opencv.org):
- looking for sponsors for OpenCV event in May in San Jose.
- recorded video w. OpenMV team to help their campaign.
- bin picking challenge is going on.
- talking to existing members of opencv.org regarding collaboration in 2025.
- Embedded Vision Summit in Santa Clara is coming, going to participate there as well.

2025-02-26

Alexander Smorkolov (opencv.ai):
- one more person, Ivan, will join OpenCV team as part-time developer. He will be working on improving OpenCV CI.
- finally finished regular 4.x=>5.x merge (PR #26939)
- working on RVV HAL, a few PRs have been merged.
- big design question is how external HAL will handle parallel loops. Preferrably, HAL should reuse OpenCV cv::parallel_for_() framework in order to reuse the same thread pool.
- talking to ARM regarding expanding KleidiCV-based HAL to support Linux as well (currently, it only supports Android).
- thinking of moving all HAL implementations from opencv/3rdparty into a dedicated directory opencv/hal.
- want to design HAL in such a way that different HAL implementations may co-exist and be called for different use cases.
Yuantao Feng (OpenCV China):
- vectorized cv::norm() for 1 array using universal and native RVV intrinsics. PR #26885 was submitted and merged.
- vectorized cv::norm() for 2 arrays (difference norm), PR will be submitted shortly.
- tried to optimize Mat::copyTo() with mask with partially successful results. On RVV universal intrinsics are not good enough/fast enough for this kernel, but native intrinsics helped to accelerate it.
- thinking about moving RVV-specific kernels into HAL.
- working on improved Aruco detection that supports different dictionaries at once.
Shiqi:
- have some students in sustech lab working on RVV optimizations.
- some more students may work on enabling other camera backends, e.g. for raspberry pi.
- talking to Orbbec regarding possible further collaboration.
Gursimar:
- working on MCC detector PR, cleaning up code.
Gary:
- [Update: we are in!] Tomorrow Google will decide on accepted orgs for GSoC 2025. If we are accepted:
  - would like to continue working on spatial CV.
  - DISK feature detector+descriptor (+lightglue matcher) will be very useful for different applications, might be useful to implement it as GSoC 2025 project.

2025-02-19

Alexander Smorkolov (opencv.ai):
- Finalized all open PRs with OpenVX integrations ## 26903, 26911, 26907. Submitted OpenVX issue #58.
- Going to adopt the new CPU HAL mechanism introduced by KleidiCV team that will use less macros, less code per each function.
- Experimenting with Aruco markers for navigation.
- Prepared large 4.x -> 5.x merge #26939. That's quite a big patch.
- Deployed RISC-V RVV build and test infrastructure locally. Reviewed, benchmarked ## 26941, 26923, 26918, 26892, 26885, 26887, 26865. Filed bug #26936.
- Reviewed, tested, helped to resolve issues ## 26940, 26934, 26913, 26906, 26926, 26914, 26785.
Yuantao Feng (OpenCV China):
- Optimizations for SpaceMIT K1:
  - core: vectorize cv::norm: #26885. Performance regressions are fixed. Waiting to be merged.
  - core: vectorize normDiff. Finished implementation based on the above one. Will submit the PR once the base PR is merged.
  - core: vectorize cv::convertTo (mainly about converting from boolean to other data types). Vectorization does not gain performance improvement. Maybe compiler is good enough.
- imgproc: fix perf regressions of c3 kernels. Will try to finalize the PR in the following week. Vadim mentioned that we may need universal intrinsics for interpolation, e.g. cv::bilinear_interpolate(const v_uint8& r, const v_uint8& g, const v_uint8& b) and something like that.
Vincent Rabaud:
- Played with SIMD in opencv.js with emscripten. Compiles well.
- The question of building OpenCV w/o exceptions is raisen again. In imgcodecs in particular OpenCV uses exceptions inconsistently. Maybe it can be GSoC project (under the umbrella with a few similar activities, e.g. proper Mat constness).
Gursimar:
- Made 2 more changes in MCC detector. Prepared PR to move MCC detector from opencv_contrib to opencv.
Shiqi:
- New semester in Univ. has just started. Will meet with students tomorrow. Maybe some of them may wish to participate in OpenCV development.
- Q to Phil: has workshop on bin picking challenge been accepted by CVPR? A: yes!
Phil:
- ARM posted newsroom article on KleidiCV+OpenCV integration: https://newsroom.arm.com/blog/arm-kleidicv-opencv-integration.
- still preparing for the conference.
- attending show this week regarding autonomous car racing: https://www.code19.ai/
- talking to a few companies about donations to OpenCV.
Vadim:
- playing a bit with AI code assistants, hopefully it will help to accelerate DNN refactoring.

2025-02-12

Alexander Smorkolov (opencv.ai):
- working on HAL refactoring/improvements.
- now trying to get rid of "unofficial" HAL's, e.g. IPP or OpenVX, that is, acceleration libraries that are used in OpenCV directly without using intermediate universal HAL API.
- all HAL calls will be enclosed into the OpenCV standard tracing mechanism so when OpenCV is built with proper flags, all calls to HAL can be traced and collected, e.g. for companies to understand which low-level functions are used in some high-level algorithms.
- there are customers who are interested in OpenCV's RNG in Python.
Yuantao Feng (OpenCV China):
- PR with cv::transform() optimization for RISC-V+RVV has been merged.
- did expriments on cv::norm vectorization, faced some regressions.
- talked to RISC-V expert (Han Liutong) regarding slow RVV code in OpenCV when number of vlanes is unknown at compile time.
  - Alexander S: in relatively new versions of GCC (11.1 or later) there are significant improvements to generate much better ARM code; maybe, for RVV the latest versions of GCC should be checked.
Vincent Rabaud:
- nothing to disclose publicly
Gary:
- had nice time skiing
- waiting for Google to let us into GSoC 2025 or not.
- Q: are we interested to play with NVidia Orin? A from Alexander S: the goal is not quite clear without a particular customer (or maybe NVidia themselves) asking for it.
Gursimar:
- PR with inpainting sample has been merged to 5.0 branch. The model is based on diffusion model and is supported by OpenCV well (the speed is quite low though).
- created draft PR to move MCC detector into main repository.
Phil Nelson (opencv.org):
- Mostly continuing working on logistics to prepare for OpenCV SID conference.
  - prepared promotional video for it.
- there are already 1000+ submissions to the bin picking challenge.
Vadim Pisarevsky:
- no updates this week.

2025-02-05

Alexander Smorkolov (opencv.ai):
- finally merged the latest relevant changes in 4.x to 5.x
- working on incoming PRs
- started working on request from IPP team, i.e. creating placeholder in opencv/3rdparty for new isolated implementation of IPP-based OpenCV HAL. Facing some problems because sometimes interaction with IPP is integrated into OpenCV quite seamlessly, difficult to decouple.
- looks like we stuck with the solution for std::vector<>.
Yuantao Feng (OpenCV China):
- got back from Chinese New Year holidays. Submitted PR to vectorize cv::transform(). Mostly achieved good acceleration except for 64f data type.
- also working on convertTo acceleration
- will try to finalize PR to fix perf regressions in the new warping kernels: #26752.
Shiqi Yu (OpenCV China):
- Currently in Spain
- Visited Prof. Rafael Munoz Salinas, the author of ArUco; discussed the latest developments, variants/derivatives of Aruco that can provide better quality and get detectable at farther distances. Also, more artistically beautiful variations of Aruco (see QR-codes art) are considered.
Vadim Pisarevsky:
- no updates
Phil:
- still working on the conference in May.
- bin picking challenge is now officially opened: https://bpc.opencv.org. Datasets are published. 800 sign-ups already, people can send submissions.
- will participate in Embedded Vision Summit; will have CVPR workshop this year.
Gary:
- one of the ideas to pursue: effective deep features, similar to LightGlue. Need to find a solution that would have a proper license.
- another idea is to use code assistants to write opencv-based code (e.g. for samples or for 3d stuff, e.g. slam) or to write optimized code inside opencv.
Vincent:
- some fuzzing-related bugs fixed in png.
- hope to get GSoC project on making opencv more secure, especially related to image i/o.

2025-01-29

Alexander Smorkolov (opencv.ai):
- After OpenCV 4.11 has been successfully released, now working on 4.x=>5.x merge. Looks like changes in opencv_extra and contrib should be ported to 5.x as well.
- Preparing report on the current Aruco state in OpenCV. Will likely offload further Aruco development to another engineer from opencv.ai
- Talking to ARM to enable KlediCV-based acceleration on ARM Linux, not just Android. It will affect opencv_python distro for ARM Linux.
Vadim Pisarevsky:
- keep working on adding block layout support to the new dnn engine (5.0-pre):
  - added block layout-friendly maxpool, avgpool, global avgpool, depthwise conv, batchnorm as well as transform_layout layer for conversion between nchw/nhwc and block layout.
- together with Alexander met with IPP team, discussed plans on IPP integration into OpenCV 5.x through the currently designed new version of CPU HAL API. The short-term goal is to move IPP bindings to opencv/3rdparty/ipp, call most critical IPP functions via the new HAL and enable IPP warping functions (warpAffine, warpPerspective, remap).
Gursimar:
- made a few changes into the inpainting PR; PR has been approved.
- made 2 PRs with fixes in MCC (actually, one fix replicated in 4.x and 5.x branches). Approved as well.
- collecting list of dnn models that we want to enable through GSoC 2025. Made spreadsheet on Google drive.
Vincent:
- tested fresh OpenCV 4.11 internally; it's quite smooth, no problems observed.
- Q: looks like FastCV development pace is quite slow? A: indeed, FastCV functionality is quite limited and stays limited for several years already. However, some of their functions are much faster than OpenCV counterparts, so FastCV-based HAL still makes sense.
Phil:
- preparing for the event in May, signed some related papers, so it's now official.
- bin picking challenge: going to start accepting submissions on Feb 1st: https://opencv.org/blog/announcing-the-opencv-perception-challenge-for-bin-picking/
- working on collecting more funds.
- going to participate in the following event: https://lu.ma/haltvvki.
Shiqi Yu (OpenCV China):
- enjoying Chinese New Year and holidays :)

2025-01-22

Alexander Smorkolov
- Done:
  - Released OpenCV-Python and AAR packages for Android.
  - Tested and merged all PNG, JpegXL, animations related patches (## 26788, 26787, 26782, 26813). Enabled maximum imgcodecs coverage on CI (#214).
  - Made opencv-python backlog grooming.
  - Works on AVIF support in OpenCV-Python packages.
  - Reviewed, tested, helped to resolve issues with ## 26810, 26800, 26815, 26646, 26803, 26802, 26793, 26798, 26773, 26778, 26820.
- Plans:
  - Finalize 4.x->5.x merge.
  - Window resizing no longer works after 4.8.0 #26761.
  - HAL design for 5.x.
Vincent Rabaud
- Fixing leaks in PNG decoder
Suleyman Turkmen
- Was GSoC Student, Mentor
  - You can be a contributor twice
  - You can be a mentor no limit
- imread, imdecode, imdecode multi -- combine into one internal class
  - as GSoC project -- but do one example, then break into several steps
Yuantao Feng
- Working on performance regression of Warp functions
- Looking at RiscV optimizations HAL interface for OpenCV SpacemiT - RISC-V
- General RiscV HAl considerations SIMD coverage expansion
  - First pull request submitted for cvtransform -- probably for 5.x, but some can be backported to 4.x
  - If in latest 4.x, it will automatically ported to 5.x. Best to go 4.x to 5.x so that 4.x doesn't overwrite
    - This is true unless there is not support for datatypes in 4.x
  - We hope for a wiki page for how to configure that board
Gursimar Singh
- Made changes to image inpainting example and pushed
- Working on SpacialAI demo
- Color correction pull request ongoing
Shiqi Yu
- Mostly on academics
- Teaching programming, maybe on RiscV optimization
- GSOC 2025 -- Use slides from Gursimar
Phil Nelson
- Adding recordings to
- OpenCV Store (not direct on site yet)
  - To include Orbbec -- need new contact there
  - Need to put OpenCV SID Conference on Computer Vision and AI (May 12)
- Working on the marketing for SID to help advertise the Vision workshop and advert speakers
- Running bin picking challenge with intrinsic
  - Feb 1st submissions on
  - 150+ submissions so far
- YouTube 100K subscribers plaque hit!
- We will have a booth at the Embedded Vision Summit May 22
- Kat from Ros -- unofficial meetup at Photonics Jan 29th, intrinsics
Gary Bradski
- Work ongoing OpenCV SID Conference on Computer Vision and AI (May 12)
- Will go live next week
  - Big early bird specials planned
- GSoC 2025 ideas, See the Roadmap Slides
  - Camera Localization using LightGlue and SuperGlue
  - Combining imread, imdecode, imdecode multi -- combine into one internal class

To Dos

Name
- todo

2025-01-15

Yuantao F (OpenCV China):
- submitted draft PR for C3 flavors warping functions fixing regressions on Intel: https://github.com/opencv/opencv/pull/26752. Need to fix accuracy issues in CV_SIMD_SCALABLE.
- narrowed down the scope of opencv acceleration for SpaceMIT K1 RISC-V+RVV.
Vincent:
- did internal 4.11 testing; mostly good, observed just one undefined behaviour case.
- we might need 4.11.1 to provide proper patch for PNG-related security issue.
Vladimir:
- provided fix for relaxing RVV and PowerPC conditions to enable dot-product optimizations.
Gursimar:
- looking at mcc in opencv_contrib. Found a few bugs in mcc detector, some of them are fixed in https://github.com/opencv/opencv_contrib/pull/3599.
- tried color corrections and prepared little presentation about it; will discuss it in details on Friday.
Phil:
- Bin-picking challenge is now open, already got many participants (submissions will come in starting from February): https://bpc.opencv.org.
- Doing the final round of collecting funds and doing finance math about 1-day "embodied AI" conference in May.
Gary:
- just got back from CES. Met one guy from Korea, a big OpenCV fan :)
- time to start thinking about GSoC 2025:
  - some kind of SLAM would be really nice to have.
  - (input from Alexander S): better opencv for web, including rtc-based cv::videocapture, cv::videowriter for web.
  - (input from Vincent): media-pipe based web cv demos: https://mediapipe-studio.webapps.google.com/demo/image_segmenter https://ai.google.dev/edge/mediapipe/solutions/examples.
  - (input from Vincent): 1. speed, 2. security are two big directions where we can always have some projects and where Vincent can help with mentorship.
Alexander S (opencv.ai):
- Done:
  - Prepared regular 4.11.0 release on Github. Still working on OpenCV-Python 4.11.
  - Updated mac Mini M1 for OpenCV-Python builds. Handled opencv-python PRs ## 1073, 1061, 1031, 1000, 1071, 1074. Handled open tickets.
  - Prepared regular 4.x->5.x merge for opencv #26768 and opencv_contrib #3871. Stuck with very large delta in opencv_extra and related conflicts.
  - Reviewed, tested, helped to resolve issues with ## 26776, 26690, 26769, 26763, 26765, 26757, 26753, 26738, 26748, 26741, 26602, 26713, 26676, 26739, 26762, 26736, contrib ## 3868, 3867, 3866, 3870.
- Plans:
  - Android AAR release. Release notes.
  - HAL design for 5.x.
  - JpegXL related issues: memory leaks #26766 and alpha channel #26767.
  - Window resizing no longer works after 4.8.0 #26761.

2025-01-08

Vadim:
- working on adding block layout (NC1HWC0 ...) in DNN.
Yuantao F (OpenCV China):
- few updates (NY holidays). Continue working on fixing performance regressions for C3 kernels. CV_SIMD_SCALABLE branch (e.g. for RISC-V RVV) still needs to be done.
- collecting extensive performance data for ARM Cortex-A55 and RISC-V Spacemit K1 chip.
Vincent:
- Tested OpenCV 4.11-pre, everything's fine.
- std::vector refactoring is in progress, one of PRs (for 5.x) has been submitted already, some other PRs are in progress.
- thinking of adding sandbox capabilities to OpenCV: https://developers.google.com/code-sandboxing/sandbox2.
Gursimar:
- working on Lama model for inpainting. Ran ONNX model succcessfully, applied blockwise quantization that brought the model size down to 93Mb. Running time on CPU is 2.5 seconds for 512x512 image. Created PRs for opencv model zoo and opencv (new sample)
Phil:
- working on 'embodied-AI' conference (San Jose, 2025 May).
- bin-picking challenge still needs to be announced (hopefully, it will happen early next week).
- start pinging OpenCV sponsors on continue OpenCV sponsorship in 2025.
- will be attending Photonix event in US this month, possibly talking to more possible OpenCV sponsors.
Alexander S (opencv.ai):
- Almost finished preparations for OpenCV 4.11.0. Big thanks to Vincent and Google team who helped to find many issues in image codecs.
- Merged all PRs related animated image support (GIF, WebP, Avif, Png, ...)
- Updated KleidiCV-based OpenCV acceleration (via OpenCV HAL).
- For several coming months Alexander S will be the only person from opencv.ai working on OpenCV fulltime, Abduragheem may help.
- Going to revise the list of pending PRs, e.g. related to Charuco.