imgproc: fix unaligned memory access in filters and Gaussian blur #25364
* filter/SIMD: removed parts which casted 8u pointers to int causing unaligned memory access on RISC-V platform.
* GaussianBlur/fixed_point: replaced casts from s16 to u32 with union operations
Performance comparison:
- [x] check performance on x86_64 - (4 threads, `-DCPU_BASELINE=AVX2`, GCC 11.4, Ubuntu 22) - [report_imgproc_x86_64.ods](https://github.com/opencv/opencv/files/14904702/report_x86_64.ods)
- [x] check performance on AArch64 - (4 cores of RK3588, GCC 11.4 aarch64, Raspbian) - [report_imgproc_aarch64.ods](https://github.com/opencv/opencv/files/14908437/report_aarch64.ods)
Note: for some reason my performance results are quite unstable, unaffected functions show speedups and slowdowns in many cases. Filter2D and GaussianBlur seem to be OK.
Slightly related PR: https://github.com/opencv/ci-gha-workflow/pull/165
Reworked findContours to reduce C-API usage #25146
What is done:
* rewritten `findContours` and `icvApproximateChainTC89` using C++ data structures
* extracted LINK_RUNS mode to separate new public functions - `findContoursLinkRuns` (it uses completely different algorithm)
* ~added new public `cv::approximateChainTC89`~ - **❌ decided to hide it**
* enabled chain code output (method = 0, no public enum value for this in C++ yet)
* kept old function as `findContours_old` (exported, but not exposed to user)
* added more tests for findContours (`test_contours_new.cpp`), some tests compare results of old function with new one. Following tests have been added:
* contours of random rectangle
* contours of many small (1-2px) blobs
* contours of random noise
* backport of old accuracy test
* separate test for LINK RUNS variant
What is left to be done (can be done now or later):
* improve tests:
* some tests have limited verification (e.g. only verify contour sizes)
* perhaps reference data can be collected and stored
* maybe more test variants can be added (?)
* add enum value for chain code output and a method of returning starting points (e.g. first 8 elements of returned `vector<uchar>` can represent 2 int point coordinates)
* add documentation for new functions - **✔️ DONE**
* check and improve performance (my experiment showed 0.7x-1.1x some time ago)
* remove old functions completely (?)
* change contour return order (BFS) or allow to select it (?)
* return result tree as-is (?) (new data structures should be exposed, bindings should adapt)
core: doc: add note for countNonZero, hasNonZero and findNonZero #25356Close#25345
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
[BugFix] dnn (ONNX): Foce dropping constant inputs in parseClip if they are shared #25319
Resolves https://github.com/opencv/opencv/issues/25278
Merge with https://github.com/opencv/opencv_extra/pull/1165
In Gold-YOLO ,`Div` has a constant input `B=6` which is then parsed into a `Const` layer in the ONNX importer, but `Clip` also has the shared constant input `max=6` which is already a `Const` layer and then connected to `Elementwise` layer. This should not happen because in the `forward()` of `Elementwise` layer, the legacy code goes through and apply activation to each input. More details on https://github.com/opencv/opencv/issues/25278#issuecomment-2032199630.
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Ownership check in TFLite importer #25312
### Pull Request Readiness Checklist
resolves https://github.com/opencv/opencv/issues/25310
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Optimize int8 layers in DNN modules by using RISC-V Vector intrinsic. #25230
This patch optimize 3 functions in the int8 layer by using RVV Native Intrinsic.
This patch was tested on QEMU using VLEN=128 and VLEN=256 on `./bin/opencv_test_dnn --gtest_filter="*Int8*"`;
On the real device (k230, VLEN=128), `EfficientDet_int8` in `opencv_perf_dnn` showed a performance improvement of 1.46x.
| Name of Test | Original | optimized | Speed-up |
| ------------------------------------------ | -------- | ---------- | -------- |
| EfficientDet_int8::DNNTestNetwork::OCV/CPU | 2843.467 | 1947.013 | 1.46 |
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [ ] I agree to contribute to the project under Apache 2 License.
- [ ] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [ ] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
imgcodecs: jpeg: re-support to read CMYK Jpeg #25280Close#25274
OpenCV Extra: https://github.com/opencv/opencv_extra/pull/1163
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Merge with https://github.com/opencv/opencv_extra/pull/1158
Todo:
- [x] Fix Attention pattern recognition.
- [x] Handle other backends.
Benchmark:
"VIT_B_32 OCV/CPU", M1, results in milliseconds.
| Model | 4.x | This PR |
| - | - | - |
| VIT_B_32 OCV/CPU | 87.66 | **83.83** |
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Orbbec Camera supports MacOS，Gemini2 and Gemini2L support Y16 format #24877
note:
1.Gemini2 and Gemini2L must use the latest firmware -- https://github.com/orbbec/OrbbecFirmware;
2.Administrator privileges are necessary to run on MacOS.
Add imread #24415
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [X] I agree to contribute to the project under Apache 2 License.
- [X] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [X] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Hello everyone,
I created this new version of the imread function and I think it can be very useful in several cases.
It is actually passed to it object on which you want to upload the image.
The advantages can be different like in case one needs to open several large images all the same in sequence.
one can use the same pointer and the system would not allocate memory each time.
libjpeg upgrade to version 9f #25092
Upgrade libjpeg dependency from version 9d to 9f.
- [X] I agree to contribute to the project under Apache 2 License.
- [X] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [X] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
The parallel code works out how many CPUs are on the system by checking
the quota it has been assigned in the Linux cgroup. The existing code
works under cgroups v1 but the file structure changed in cgroups v2.
From [1]:
"cpu.cfs_quota_us" and "cpu.cfs_period_us" are replaced by "cpu.max"
which contains both quota and period.
This commit add support to parallel so it will read from the cgroups v2
location. v1 support is still retained.
Resolves#25284
[1] 0d5936344f
Speed up adaptive threshold in findChessboardCorners #25177
### Pull Request Readiness Checklist
If `block_size` hasn't been changed between iterations for same `k`, then all `adaptiveThreshold` arguments will be same and we can reuse result from previous iteration.
I tested this PR with benchmark
```
python3 objdetect_benchmark.py --configuration=generate_run --board_x=7 --path=res_chessboard --synthetic_object=chessboard
```
PR speed up chessboards detection by `7.5/17%` without any changes in detected chessboards number:
```
cell_img_size = 100 (default)
before
category detected chessboard total detected chessboard total chessboard average detected error chessboard
all 0.904167 13020 14400 0.600512
Total detected time: 107.27875600000003 sec
after
category detected chessboard total detected chessboard total chessboard average detected error chessboard
all 0.904167 13020 14400 0.600512
Total detected time: 99.0223499999999 sec
----------------------------------------------------------------------------------------------------------------------------------------------
cell_img_size = 10
before
category detected chessboard total detected chessboard total chessboard average detected error chessboard
all 0.539792 7773 14400 4.209964
Total detected time: 2.989205999999999 sec
after
category detected chessboard total detected chessboard total chessboard average detected error chessboard
all 0.539792 7773 14400 4.209964
Total detected time: 2.4802350000000013 sec
```
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
Added in-place support for cartToPolar and polarToCart #24893
- a fused hal::cartToPolar[32|64]f() is used instead of sequential hal::magnitude[32|64]f/hal::fastAtan[32|64]f
- ipp_polarToCart is skipped for in-place processing (it seems not to support it correctly)
relates to #24891
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [X] I agree to contribute to the project under Apache 2 License.
- [X] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [X] The PR is proposed to the proper branch
- [X] There is a reference to the original bug report and related work
- [X] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
dnn: avoid const layer forwarding in layer norm layer and attention layer #25238
While profiling ViTs with dnn, I found `ConstLayer` can take a proportion of the inference time, which is weird. This comes from the data copy during the inference of `ConstLayer`. There is a chance that we can improve the efficiency of data copying but the easiest and most convenient way is to avoid `ConstLayer`. This PR change the way how we handle constants in layer normalization layer and attention layer, which is storing in the layer blobs instead of making constant layers for them.
Checklists:
- [x] Backend compatibility in layer normalization layer.
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
doc: add note on handling of spaces in CommandLineParser #25237
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Added note that this class will not work properly if tabs and other whitespace characters are included in the key.
The support of whitespace characters by istringstream, etc. is on hold because the future of this class is not clear compared to implementations in Python and other languages.
dnn (CANN): Fix incorrect shape of 1d bias in Gemm #25166
Gemm layer was refactored some time ago. Users found that the mobilenet example in https://github.com/opencv/opencv/wiki/Huawei-CANN-Backend does not work because of incorrect shape set for 1d bias in Gemm. This PR resolves this issue.
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Release convolution weightsMat after usage #25181
### Pull Request Readiness Checklist
related (but not resolved): https://github.com/opencv/opencv/issues/24134
Minor memory footprint improvement. Also, adds a test for VmHWM.
RAM top memory usage (-230MB)
| YOLOv3 (237MB file) | 4.x | PR |
|---------------------|---------|---------|
| no winograd | 808 MB | 581 MB |
| winograd | 1985 MB | 1750 MB |
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Normaly, we sets IMWRITE_* flags for imwrite() params.
But imgcodecs expects to use some TIFFTAG_* directory.
This patch introduce IMWRITE_TIFF_ROWSPERSTRIP and
IMWRITE_TIFF_PREDICTOR instead of TIFFTAG_*.
* libtiff upgrade to version 4.6.0
* fix tiffvers.h cmake generation
* temp: force build 3rd party deps from source
* remove libport.h and spintf.c
* cmake fixes
* don't use tiff_dummy_namespace on windows
* introduce numeric_types namespace alias
* include cstdint
* uint16_t is not a numeric_types type
* fix uint16 and uint32 type defs
* use standard c++ types
* remove unused files
* remove more unused files
* revert build 3rd party code from source
---------
Co-authored-by: Misha Klatis <misha.klatis@autodesk.com>
Maksim Shabunin
Alexander Smorkalov
Kumataro
ecchen
Alexander Lyulkov
Tsukasa Sugiura
Liutong HAN
Dmitry Kurtaev
Yuantao Feng
catree
sepperliu
Yuantao Feng
zzuliys
thewoz
Michael Klatis
John Slade
Maxim Smolskiy
Pierre Chatelier
Yusuke Kameda
RoboSchmied
Alex
lpylpy0514
Carlos Azevedo
Oleg Pipikin
Misha Klatis