WeChatQR-fix conversion Caffe to ONNX #28746
### 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
Temporarily disabled RVV intrinsics in several layers of the new DNN tested on musebook K1.
Kernels will be re-enabled when we find some good solution for the current 'm2' issue in rvv_scalable intrinsics.
This should fix#28852
### 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
Update default YuNet model to new dynamic inputs #29107
Update the default model in `face_detect.py` and `face_detect.cpp` to
`face_detection_yunet_2026may.onnx`, which has symbolic `height`/`width` input dims.
## Changes
- `samples/dnn/face_detect.py`: update default `--face_detection_model` to `face_detection_yunet_2026may.onnx`
- `samples/dnn/face_detect.cpp`: update default `fd_model` to `face_detection_yunet_2026may.onnx`
Companion PR :
- https://github.com/opencv/opencv_zoo/pull/310
- https://github.com/opencv/opencv_extra/pull/1373
Closes : https://github.com/opencv/opencv/issues/28769
### 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
imgcodecs: fix Apple conversions build without AVFoundation #29138Fixes#28553.
This PR fixes Apple imgcodecs conversion builds when AVFoundation is disabled or unavailable on older macOS SDKs.
Changes:
- Guard AVFoundation import with HAVE_AVFOUNDATION.
- Use older macOS-compatible framework imports when ImageIO is unavailable.
- Add __bridge compatibility for older Objective-C++ compilers.
- Use NSMakeSize for older macOS instead of passing CGSize to NSImage API.
Testing:
- Not tested locally: no macOS environment available.
- Patch follows the fix confirmed in #28555 discussion.
Fixing python wrapper around ECCParameters to include itersPerLevel #29148
### 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
Moved geometry transformations from imgproc to 3d, future geometry module #29101
The first step of 2d geometry operations migration to the future geometry module.
I created 2d.hpp to isolate the moved functions for now. I propose to create geometry.hpp when the module is renamed and include all things there.
OpenCV contrib: https://github.com/opencv/opencv_contrib/pull/4126
### 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
- [ ] 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
### 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
- [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
<!-- Note!!! If you are an automated agent, we have a special process for you: add 🤖🤖🤖 to the end of the PR title. -->
### Summary
This PR improves SIMD coverage for masked `cv::accumulate()` on 4-channel images.
The existing masked accumulate SIMD implementation has special paths for `cn == 1` and `cn == 3`, while 4-channel images fall back to the general implementation. This PR adds `cn == 4` SIMD paths using OpenCV Universal Intrinsics.
Compare link:
https://github.com/opencv/opencv/compare/4.x...feitianduowen:opencv:4.x
Modified files:
- `modules/imgproc/src/accum.simd.hpp`
- Add masked `cn == 4` SIMD paths for:
- `CV_32FC4 -> CV_32FC4`
- `CV_8UC4 -> CV_32FC4`
- `modules/video/test/test_accum.cpp`
- Add deterministic correctness regression tests for the new masked 4-channel accumulate paths.
- `modules/imgproc/perf/perf_accumulate.cpp`
- Add performance coverage for masked 4-channel accumulate.
imgproc: add SIMD paths for masked 4-channel accumulate #29094
### Implementation
The new SIMD branches are added in:
- `modules/imgproc/src/accum.simd.hpp`
They handle:
- `src`: `CV_32FC4`, `dst`: `CV_32FC4`, `mask`: `CV_8UC1`
- `src`: `CV_8UC4`, `dst`: `CV_32FC4`, `mask`: `CV_8UC1`
For the `CV_32FC4` path, the implementation uses `v_load_deinterleave`, applies the pixel mask to all 4 channels, accumulates into `dst`, and stores the result with `v_store_interleave`.
For the `CV_8UC4 -> CV_32FC4` path, the implementation loads 4 interleaved uchar channels, applies the pixel mask, widens the values to float, accumulates into `dst`, and stores the results back as interleaved `CV_32FC4`.
The masked SIMD block is guarded with:
```cpp
if (x <= len - cVectorWidth)
{
...
}
```
This check ensures that SIMD setup and vector operations are only executed when at least one full vector iteration can run. For very small inputs or tail-only cases, the function skips SIMD initialization and lets the existing scalar fallback handle the data through `acc_general_(src, dst, mask, len, cn, x)`.
This keeps tail handling unchanged and avoids unnecessary SIMD initialization when the vector loop would be skipped.
### Accuracy tests
Added a deterministic regression test in:
* `modules/video/test/test_accum.cpp`
New test:
```bash
Video_Acc.accuracy_32FC4_masked_cn4
Video_Acc.accuracy_8UC4_to_32FC4_masked_cn4
```
Test commands:
```bash
opencv_test_video --gtest_filter=Video_Acc.accuracy_32FC4_masked_cn4
opencv_test_video --gtest_filter=Video_Acc.accuracy_8UC4_to_32FC4_masked_cn4
opencv_test_video --gtest_filter="Video_Acc.*:Video_AccSquared.*:Video_AccProduct.*:Video_RunningAvg.*"
```
The filtered accumulate-related tests passed locally.
I added dedicated deterministic tests instead of only extending the existing randomized accumulate tests because this PR only changes the masked `cv::accumulate()` paths for specific 4-channel type combinations. It does not add `cn == 4` SIMD coverage for `accumulateSquare`, `accumulateProduct`, or `accumulateWeighted`.
The existing base accumulate tests are shared by several accumulation functions. Extending the common randomized channel selection to include `cn == 4` would also affect tests for functions that are not optimized by this PR. The new deterministic tests directly cover the modified paths:
* `cv::accumulate(src, dst, mask)`
* `CV_32FC4 -> CV_32FC4`
* `CV_8UC4 -> CV_32FC4`
* `mask`: `CV_8UC1`
The test also covers small sizes and non-vector-multiple sizes, so both the new SIMD path and the existing scalar tail fallback are exercised.
### Performance tests
Added performance coverage in:
* `modules/imgproc/perf/perf_accumulate.cpp`
Perf command:
```bash
opencv_perf_imgproc --gtest_filter="*AccumulateMask32FC4*" --perf_min_samples=1000 --perf_force_samples=1000
opencv_perf_imgproc --gtest_filter="*AccumulateMask8UC4To32FC4*" --perf_min_samples=1000 --perf_force_samples=1000
```
Test environment:
* Release build
* AVX2 enabled
* IPP disabled
* OpenCL disabled
* Windows MinGW build
Performance results:
`CV_32FC4 -> CV_32FC4`
| Size | Before median | After median | Speedup |
| --------- | ------------- | ------------ | ------- |
| 1920x1080 | 3.07 ms | 2.74 ms | 1.12x |
| 1280x720 | 0.84 ms | 0.73 ms | 1.15x |
| 640x480 | 0.17 ms | 0.16 ms | 1.06x |
| 320x240 | 0.04 ms | 0.04 ms | ~1.00x |
`CV_8UC4 -> CV_32FC4`
| Size | Before median | After median | Speedup |
| --------- | ------------- | ------------ | ------- |
| 1920x1080 | 6.09 ms | 1.75 ms | 3.48x |
| 1280x720 | 2.69 ms | 0.82 ms | 3.28x |
| 640x480 | 0.96 ms | 0.28 ms | 3.43x |
| 320x240 | 0.24 ms | 0.06 ms | 4.00x |
| 127x61 | 0.02 ms | 0.01 ms | 2.00x |
### Build and test commands
The following commands were run from Windows `cmd.exe`.
```sh
cmake -S . -B build_release -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTS=ON -DBUILD_PERF_TESTS=OFF -DBUILD_EXAMPLES=OFF -DWITH_IPP=OFF -DWITH_OPENCL=OFF -DCMAKE_C_FLAGS=-mstackrealign -DCMAKE_CXX_FLAGS=-mstackrealign
cmake --build build_release --target opencv_test_video -j 4
build_release\bin\opencv_test_video.exe --gtest_filter=Video_Acc.accuracy_32FC4_masked_cn4
build_release\bin\opencv_test_video.exe --gtest_filter=Video_Acc.accuracy_8UC4_to_32FC4_masked_cn4
build_release\bin\opencv_test_video.exe --gtest_filter=Video_Acc.*:Video_AccSquared.*:Video_AccProduct.*:Video_RunningAvg.*
```
Accuracy test commands passed locally.
For the performance comparison, I kept the new perf test in `modules/imgproc/perf/perf_accumulate.cpp` and only reverted `modules/imgproc/src/accum.simd.hpp` to measure the baseline. Then I restored the SIMD patch and measured the optimized version.
Baseline measurement:
```sh
git diff -- modules/imgproc/src/accum.simd.hpp > accum_cn4_simd.patch
git checkout -- modules/imgproc/src/accum.simd.hpp
mkdir perf_logs
cmake -S . -B build_perf_before -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTS=OFF -DBUILD_PERF_TESTS=ON -DBUILD_EXAMPLES=OFF -DWITH_IPP=OFF -DWITH_OPENCL=OFF -DCMAKE_C_FLAGS=-mstackrealign -DCMAKE_CXX_FLAGS=-mstackrealign > perf_logs\before_cmake_config.txt 2>&1
cmake --build build_perf_before --target opencv_perf_imgproc -j 4 > perf_logs\before_build.txt 2>&1
build_perf_before\bin\opencv_perf_imgproc.exe --gtest_filter=*AccumulateMask32FC4* --perf_min_samples=1000 --perf_force_samples=1000 > perf_logs\before_perf.txt 2>&1
```
```sh
git diff -- modules/imgproc/src/accum.simd.hpp > accum_cn4_u8_simd.patch
git checkout -- modules/imgproc/src/accum.simd.hpp
cmake -S . -B build_perf_before_u8 -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTS=OFF -DBUILD_PERF_TESTS=ON -DBUILD_EXAMPLES=OFF -DWITH_IPP=OFF -DWITH_OPENCL=OFF -DCMAKE_C_FLAGS=-mstackrealign -DCMAKE_CXX_FLAGS=-mstackrealign
cmake --build build_perf_before_u8 --target opencv_perf_imgproc -j 4
build_perf_before_u8\bin\opencv_perf_imgproc.exe --gtest_filter=*AccumulateMask8UC4To32FC4* --perf_min_samples=1000 --perf_force_samples=1000 > perf_logs\before_perf1.txt 2>&1
```
Optimized measurement:
```sh
git apply accum_cn4_simd.patch
cmake -S . -B build_perf_after -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTS=OFF -DBUILD_PERF_TESTS=ON -DBUILD_EXAMPLES=OFF -DWITH_IPP=OFF -DWITH_OPENCL=OFF -DCMAKE_C_FLAGS=-mstackrealign -DCMAKE_CXX_FLAGS=-mstackrealign > perf_logs\after_cmake_config.txt 2>&1
cmake --build build_perf_after --target opencv_perf_imgproc -j 4 > perf_logs\after_build.txt 2>&1
build_perf_after\bin\opencv_perf_imgproc.exe --gtest_filter=*AccumulateMask32FC4* --perf_min_samples=500 --perf_force_samples=500 > perf_logs\after_perf.txt 2>&1
```
```sh
git apply accum_cn4_u8_simd.patch
cmake -S . -B build_perf_after_u8 -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTS=OFF -DBUILD_PERF_TESTS=ON -DBUILD_EXAMPLES=OFF -DWITH_IPP=OFF -DWITH_OPENCL=OFF -DCMAKE_C_FLAGS=-mstackrealign -DCMAKE_CXX_FLAGS=-mstackrealign
cmake --build build_perf_after_u8 --target opencv_perf_imgproc -j 4
build_perf_after_u8\bin\opencv_perf_imgproc.exe --gtest_filter=*AccumulateMask8UC4To32FC4* --perf_min_samples=1000 --perf_force_samples=1000 > perf_logs\after_perf1.txt 2>&1
```
Performance comparison was measured by keeping the new perf tests and only reverting modules/imgproc/src/accum.simd.hpp for the baseline run.
`-mstackrealign` was used for the MinGW Windows build to avoid stack-alignment issues with AVX code generation during local testing.
### Notes
The implementation keeps the existing scalar fallback path unchanged. SIMD is used only for the newly covered masked `cn == 4` vectorizable part, and any remaining tail elements are still handled by `acc_general_()`.
The improvement is most visible on larger images. Small images are dominated by overhead and do not always show meaningful speedup.
Added SDPA layer (Scaled Dot Product Attention) #29104
Merge with: https://github.com/opencv/opencv_extra/pull/1374
### 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
Migrated chessboard and circles grid detectors to objdetect #28804
OpenCV Contrib: https://github.com/opencv/opencv_contrib/pull/4125
OpenCV Extra: https://github.com/opencv/opencv_extra/pull/1375
### 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
- [ ] 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
fix MSVC error: use name constexp for alignas array size in INT8 kernel #29156
### Problem
Windows CI fails with MSVC error C2131 in `conv2_int8_kernels.simd.hpp`:
error C2131: expression did not evaluate to a constant
failure was caused by a read of a variable outside its lifetime
see usage of 'this'
Two array declarations inside `parallel_for_` lambdas used `constexpr`
variables defined inside the lambda body as array sizes:
alignas(32) int8_t wbuf[128 * K0]; // K0 defined inside lambda
alignas(32) int32_t sumbuf[SPAT_BLOCK_SIZE * K0]; // both defined inside lambda
### Fix
Move the combined size constants to function scope (before the lambda),
where they are true compile-time constants with no `this` involvement:
constexpr int WBUF_SIZE = 128 * 8; // 128 * K0
constexpr int SUMBUF_SIZE = 8 * 8; // SPAT_BLOCK_SIZE * K0
### Related
Fixes Windows CI failure introduced by #29126.
### 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
<!-- Note!!! If you are an automated agent, we have a special process for you: add 🤖🤖🤖 to the end of the PR title. -->
Attention graph fusion with MLAS FlashAttention #29126
Performance numbers for Owl-v2 model on intel i9:
```
ORT: Average inference time over 10 runs: 1411.55 ms (min 1399.75, max 1438.89)
NEW: Average inference time over 10 runs: 1078 ms (min 1048.04, max 1110.61)
```
### 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
Casting the raw integer fields ras_type and maptype directly to their
respective enum types (SunRasType / SunRasMapType) without a prior range
check is undefined behavior under C++11 §7.2/8 when the stored value falls
outside the enum's valid range.
UBSan reports:
grfmt_sunras.cpp:72: runtime error: load of value 34077, which is not a
valid value for type 'SunRasMapType'
Fix: read both fields into plain ints first, validate them against the
declared enumerator bounds, and return false (reject the image) on any
out-of-range value before performing the enum cast. This is the cheapest
correct approach — two integer comparisons added to a path that was already
doing I/O — and ensures no downstream code ever sees an ill-formed enum
value.
Add test_sunraster.cpp with regression tests covering:
- The exact crash_001 payload (invalid maptype 34077 / 0x851d)
- Invalid ras_type values
- maptype = 2 and UINT_MAX (outside [0,1])
- Truncated header
- A valid 8-bpp grayscale image that must still decode correctly
Signed-off-by: FuzzAnything fuzzanything@gmail.com
Merge pull request #29134 from svigh/gapi_u8_nd_mats_onnx_layout_fix
Gapi u8 N-D mats onnx layout workaround #29134
### 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
- [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
core: fix numerical instability and UB in VBLAS SIMD helpers #29140
### Description
**Fixes #28845**
**Root cause identified by:** stubbing out `VBLAS<double>::givens()` to return 0 immediately — both tests pass, confirming the divergence originates in the double-precision Givens SIMD rotation.
**Main fix:** Use FMA form in `VBLAS::{float,double}::givens()` to keep JacobiSVD vector rotation numerically stable on SIMD/FMA builds.
This fixes the fisheye homography initialization divergence (`RMS 36.2553` vs `1`) seen in `RegisterCamerasTest.hetero1/2` under GCC 13.
**Additional fix:** Replace fixed-size temporary storage `sbuf[2]` in `VBLAS<double>::dot()` with `v_reduce_sum()` to avoid out-of-bounds stores on wide SIMD backends (e.g., AVX2 `v_float64` requires 4 lanes, causing UB).
**Verification:**
Passed full `opencv_test_core` and `opencv_test_calib` locally under Ubuntu 24.04 (GCC 13) with AVX2.
### 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 (`next`)
- [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 (Passed existing `opencv_test_core` and `opencv_test_calib` tests)
- [ ] The feature is well documented and sample code can be built with the project CMake (N/A - Bug fix)
dnn int8 optimization #29079
all_layers.hpp
- Add float_input flag to Conv2Int8Params and Conv2Int8Layer to let the first conv accept raw FP32 input and quantize internally.
graph_fusion_qdq.cpp :
- Fuse DQ → Sigmoid → QL into SigmoidInt8, Similarly for MAxPool.
- Fuse the input QuantizeLinear node into the first Conv2Int8.
conv2_int8_layer.cpp
- Add quantizeInterleaveBlock()
conv2_int8_kernels.simd.hpp
- Add spatial tiling to both convInt8BlockVNNI and convInt8BlockDepthwise: splits output pixels into tiles so total task count is N × ngroups × Kblk × ntiles, fully utilizing all threads even when the channel count is small.
elementwise_layers.cpp
- Widen CV_Assert to accept CV_8U in addition to CV_8S.
eltwise2_int8_layer.cpp
- Add QLinearMul support: new Mul math path for both signed and unsigned int8.
- Add numpy-style broadcast support so QLinearMul / QLinearAdd with scalar
### 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
Add KV cache with paged attention and prefetch #29127
Closes: https://github.com/opencv/opencv/issues/27159
### 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
FP32 KV Cache #28840
OpenCV Extra: https://github.com/opencv/opencv_extra/pull/1348
## This PR introduces basic (Paged ) KV-Cache to use on CPU
### Summary:
1. To ensure proper gemm-prepacking,
1.1. The Page Size of Key Cache is currently hardcoded as `FAST_GEMM_F32_NR`(which is 8, 12 or 16 depending on CPU architecture)
1.2. The Page Size of Values Cache is hardcoded as `FAST_GEMM_F32_PACKED_STRIDE_K`
2. there are two phases supported - prefill & generate.
2.1. prefill grows cache by `N` tokens and is allowed **only** for empty cache
2.2. generate grows cache by 1 token.
2.3. **Improtant**: it is currently not allowed to grow non-empty cache by more than one token at a time (thisbehaviour is sufficient for normal LLM querying, but should be extended if we want to implement speculative decoding)
### 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
- [ ] 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
Dnn pow layer output type fix#29124
### Description
This pull request needs https://github.com/opencv/opencv_extra/pull/1372 to run.
**issue:** when running disk-lightglue `.onnx` model, such error message is thrown:
```bash
Error message OpenCV(5.0.0-pre) /home/tapakya/cpp/opencv_mod/opencv/modules/dnn/src/layers/nary_eltwise_layers.cpp:410: error: (-2:Unspecified error) in function 'virtual void cv::dnn::NaryEltwiseLayerImpl::getTypes(const std::vector<int>&, int, int, std::vector<int>&, std::vector<int>&) const'
> All inputs should have equal types (expected: 'inputs[0] == input'), where
> 'inputs[0]' is 11 (CV_64SC1)
> must be equal to
> 'input' is 5 (CV_32FC1)
```
This is because in `nary_eltwise_layers.cpp`, `getTypes` didn't set the output type of power operation correctly. `int`^`float` should output `float`, whereas `getTypes` set the output type to `int` in this case.
```c++
if (op == OPERATION::POW) {
CV_Assert(inputs.size() == 2);
auto isIntegerType = [](int t) {
return t == CV_8S || t == CV_8U || t == CV_16S || t == CV_16U || t == CV_32S || t == CV_32U || t == CV_64S || t == CV_64U;
};
auto isFloatType = [](int t) {
return t == CV_32F || t == CV_64F || t == CV_16F || t == CV_16BF;
};
int out_type;
const bool baseIsInt = isIntegerType(inputs[0]);
const bool expIsInt = isIntegerType(inputs[1]);
const bool baseIsFloat = isFloatType(inputs[0]);
const bool expIsFloat = isFloatType(inputs[1]);
if ((baseIsInt && expIsInt) || (baseIsFloat && expIsFloat))
{
out_type = (inputs[0] == inputs[1]) ? inputs[0] : CV_32F;
}
else if (baseIsFloat != expIsFloat)
{
out_type = inputs[0];
// if base is int and exp is float, output type should be float
}
```
**fix:** corrected the logic of determining output type
```c++
if ((baseIsInt && expIsInt) || (baseIsFloat && expIsFloat))
{
out_type = (inputs[0] == inputs[1]) ? inputs[0] : CV_32F;
}
else if (baseIsFloat && !expIsFloat)
{
out_type = inputs[0];
}
else if (!baseIsFloat && expIsFloat)
{
out_type = CV_32F;
}
```
### 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
- [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: skip Conv2Int8 fusion for grouped convs with Kg<8 (#28798) #28920
OpenCV Extra: https://github.com/opencv/opencv_extra/pull/1360
- [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
---
Fixes#28798.
The YuNet 2023mar int8 model from opencv_zoo doesn't detect anything on 5.x. With `--score_threshold
0.05` the detector still returns no boxes, so it's not just a confidence drop — the network output is
broken.
After dumping intermediate tensors and comparing against onnxruntime, the `obj_*` branches collapse to
~0 (saturated int8 `-128`), and the `cls_*` branches saturate the other way. Final score is `cls * obj`
so nothing ever crosses threshold.
The cause is in `Conv2Int8`'s VNNI kernel. It processes `K0 = 8` output channels per SIMD iteration and
writes them as one `K0`-wide block to `out + (n*K1 + k1)*planesize` with `k1 = k_base / K0`. When
`ngroups > 1` and `Kg = K/ngroups < K0`, consecutive groups share the same `k1` slot and overwrite each
other — only the last group's result survives. YuNet has six `1x1x3x3` depthwise conv heads (`Kg = 1`),
which is exactly this case.
The unfused `DequantizeLinear → Conv2 → QuantizeLinear` path is fine. The minimal fix here is to skip
the rewrite to `Conv2Int8` when `Kg < 8` so we fall back to the float path. A proper depthwise int8
kernel can be added later as a separate optimization.
Verified with `samples/dnn/face_detect.py` on Lena:
Before:
AssertionError: Cannot find a face in samples/data/lena.jpg
After:
Face 0, top-left coordinates: (204, 187), box width: 149, box height 212, score: 0.89
Cross-check vs onnxruntime: `cls_8` mean `0.673` (was `0.93`, ORT `0.673`), `obj_8` max `0.0039` (was
`0`, ORT `0.0039`).
The regression test is a single 16-group depthwise `QLinearConv` with non-zero `x_zp`, added to
`Quantized_Convolution`. Test data is in opencv_extra on the matching branch (~9 KB total).
<img width="1864" height="1060" alt="image" src="https://github.com/user-attachments/assets/a1b15c64-6b84-42b1-86a4-1d55474cb38c" />
Fixed write return status in videoio module#29087
closes: https://github.com/opencv/opencv/issues/24287
### 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
Alexander Smorkalov
Alexander Smorkalov
omrope79
Vadim Pisarevsky
Varun Jaiswal
vrooomy
Licardo_du
4ekmah
Vibhor
Siyu Wang
Vincent Rabaud
Abhishek Gola
example
Vigh Sebastian
Zixi Wang
Prasad Ayush Kumar
Fulnergy
nklskyoy
Shelia J.
sujal