diff --git a/2022.md b/2022.md index 10744cf..2fd345b 100644 --- a/2022.md +++ b/2022.md @@ -33,6 +33,48 @@ ## 2022-07-13 +* Vadim: + * keep working on inference engine in Ficus. Several image classification models from ONNX model zoo are now supported: Squeeznet, EfficientNetv4-lite, Shufflenet, Googlenet, Inception v1 (v2 proces incorrect results, probably incorrect preprocessing; v3 fails to load because of using ONNX sequences). Now working on FP16 mode + * GSoC: the 1st RISC-V PR is merged, now the student is preparing update to fix compile errors when all SIMD is turned off + * OpenCV China: good progress on multi-task model (object detection + pose estimation + semantical segmentation), the quality is 2% lower than claimed in the paper, but the results look reasonably good. + * OpenCV China: 2 PRs on text detection & recognition have been submitted (they contain benchmarks to evaluate quality). + * OpenCV China: 1st PR on initial Ascend support (toolchain detection for now) is submitted. + * FP16 acceleration of convolution is submitted, but fails to compile. + * AR: invite mentors for the next Wednesday (GSoC phase 1 evaluation: July 25th - July 29th). +* Alexander S (x.ai): + * Done: + * Aruco patch is ready and waiting for review (for ~2 weeks already). + * Significant progress on CI. Even Android on 3.4 branch will likely be tested (using some old NDK release version). Megapack for Windows is not ready not. + * 2 patches by Rostislav (ICP improvements) are ready for review. + * Andrey: + * Split PR docs workflows by a branch and updated docker images, 3.4 and 4.x based on Ubuntu 20.04, 5.x based on Ubuntu 22.04. [#35](https://github.com/opencv/ci-gha-workflow/pull/35), [#8](https://github.com/opencv-infrastructure/opencv-gha-dockerfile/pull/8) + * Finished fixing compiling warnings in case of hosts issues on Linux x86_64 / ARM64 and macOS x86_64 / ARM64. + * Fixed defining a path for a local variable to new DNN models on macOS. [#37](https://github.com/opencv/ci-gha-workflow/pull/37) + * Created an issue about Java Finalize Class.[ #22260](https://github.com/opencv/opencv/issues/22260) + * Renamed Ubuntu version from 1804 to 2004 for Linux ARM64 builds (just UI change). [#22266] (https://github.com/opencv/opencv/pull/22266), [#22267](https://github.com/opencv/opencv/pull/22267), [#22268](https://github.com/opencv/opencv/pull/22268) + * Fixed an issue with the recursion in opencv-python using “ipython”. [#22269](https://github.com/opencv/opencv/pull/22269) + * Alexander: + * Migrated Android environment for 3.4 to old NDK 10e. Modern NDK is not filly supported there. + * Reviewed, tested, helped to resolve issues with ## 22274, 22226, 22269, 22257, #22150, contrib #3240 + * In progress: + * Andrey: + * Fixing compiling warnings in case of hosts issues on Windows. + * Rostislav: + * Normal computer tests: draft PR #22241 filed; need to check all available data types & rewrite tests on Gtest params; add parameters to Linemod & expose CV_16U as available depth type + * 3D module review [draft](https://docs.google.com/document/u/1/d/1cJQzE5c_4N6bltw0wR1KvEAeYa9U2-wqRAyjIXikn8M/edit) written, need to add notes on API consistency & Python wrappers available +* Vincent: + * porting internal calibration tool to OpenCV 4.x, submitted some PR with fixes + * it would be nice to have an up-to-date comprehensive list of bit-exact functions somewhere +* Shiqi Yu: + * no updates so far. + + +
+<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> ++ +## 2022-07-13 + * Vadim: * Keep working on Ficus NN. MobileNet-SSD and TinyYolo v3 are now supported. The further plans include 8-bit & Winograd convolution, FasterRCNN and Mark-RCNN support (ultimately, the goal is to support every single model in ONNX model zoo and OpenCV model zoo), engine refactoring and x86-64 support. * There is good progress on Loops JIT engine.