mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-12-12 07:09:12 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Updated 2021 (markdown)

Vadim Pisarevsky 2021-12-15 18:10:03 +03:00
parent 5465a8fb3a
commit 4821c86529
1 changed files with 39 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
2021.md

@ -30,6 +30,45 @@
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
</pre>
## 2021-12-15
### *__Minutes__*
* Vadim:
* Ficus: compiler quality and the standard library have been improved; performance of the all 5 benchmarks has been improved. The speed was re-measured on Mac M1: Ficus 1.0alpha+ is on on par with C++, faster than Julia 1.7 on 3 benchmarks, slower on the other two, noticeably faster than Python+numba.
* OpenCV China:
* Performance of Tim-VX backend was measured on Khadas board, YOLO v4 is 10x faster, Shiqi Yu's YuNet is 3x faster.
* Many models from opencv model zoo have been quantized. There is some quality issue with some models, larger calibration dataset will be used.
* Alexander V:
* The fix for PnPRansac with USAC parameters has been merged: https://github.com/opencv/opencv/pull/21166
* PR about GStreamer support in Audio I/O is going to be created this week. It covers wav format only. Mp3 and media containers support will be added into dedicated PR.
* ResizeLN fix is merged (the issue happened for depth >8U/8S, if the resizeLN OpenCL kernel is compiled without INTER_LINEAR_INTEGER, it gives invalid results): https://github.com/opencv/opencv/pull/21228
* GTK2 with OpenGL compilation issue has been fixed: https://github.com/opencv/opencv/pull/21213
* Levenberg-Marquadt algorithm implementation: PR was moved from draft to ready state, some comments already received: https://github.com/opencv/opencv/pull/21018
* Anna:
* Done:
* Andrey: Tested python limited API for OpenCV python packages successfully. Modified scripts and CI/CD workflows in opencv-python repository to support python limited API. PR
* Andrey: Replaced distutils with shutil for files in 4.x branch in opencv repository which are not located in 3.4 branch. PR
* Egor: Fixed a few issues in MaxUnpool, ReduceMean/Max/Sum, Pad: fixed MaxUnpool missing attributes. fixed ReduceMean/Max/Sum keep_dims defaults and added support for keepdims=true in full reduction mode. added edge padding type to Pad.
* Egor: ONNX conformance tests PR merged.
* Egor: created OpenCV ONNX test summary table.
* Egor: reviewed Add BatchMatMul layer support for tf_importer #21154, Add Square layer support for tf_importer #21246.
* In progress:
* Andrey: Implementing download models function in a python package.
* Alexander P: Update ArUco tutorial #3126. fix “Detection of Diamond Markers” (by using custom dictionary and custom detector parameters) and “Calibrate with ArUco and ChArUco” (by new calibration photos) chapters and samples. add indices and corners check to tests:
* can_find_gboriginal
* can_find_singlemarkersoriginal
* can_find_choriginal
* can_find_diamondmarkers
now trying to pass the review.
* Alexander P: Aruco detectMarkers crashing #595. Fixed type cast error, added test. PR is being prepared.
* Alexander P: PR #20949 remap large images support. New solution to the problem was developed with Alexander Smorkalov, Vadim Pisarevsky and Alexander Alekhin.
* Egor: Made a working version in prod(broadcasting dims) calls to copyTo() in my branch. WIP on log^d(max_broad_dim) version. The next step will be evaluation and I'll probably need to optimize away mat indexation since it reallocates internal buffers in each iteration.
<pre>
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
</pre>
## 2021-12-08
### *__Minutes__*