Fix LSTM support in ONNX
* fix LSTM and add peephole support
* disable old tests
* turn lambdas into functions
* more hacks for c++98
* add assertions
* slice fixes
* backport of cuda-related fixes
* address review comments
* dnn: LSTM optimisation
This uses the AVX-optimised fastGEMM1T for matrix multiplications where available, instead of the standard cv::gemm.
fastGEMM1T is already used by the fully-connected layer. This commit involves two minor modifications:
- Use unaligned access. I don't believe this involves any performance hit in on modern CPUs (Nehalem and Bulldozer onwards) in the case where the address is actually aligned.
- Allow for weight matrices where the number of columns is not a multiple of 8.
I have not enabled AVX-512 as I don't have an AVX-512 CPU to test on.
* Fix warning about initialisation order
* Remove C++11 syntax
* Fix build when AVX(2) is not available
In this case the CV_TRY_X macros are defined to 0, rather than being undefined.
* Minor changes as requested:
- Don't check hardware support for AVX(2) when dispatch is disabled for these
- Add braces
* Fix out-of-bounds access in fully connected layer
The old tail handling in fastGEMM1T implicitly rounded vecsize up to the next multiple of 8, and the fully connected layer implements padding up to the next multiple of 8 to cope with this. The new tail handling does not round the vecsize upwards like this but it does require that the vecsize is at least 8. To adapt to the new tail handling, the fully connected layer now rounds vecsize itself at the same time as adding the padding(which makes more sense anyway).
This also means that the fully connected layer always passes a vecsize of at least 8 to fastGEMM1T, which fixes the out-of-bounds access problems.
* Improve tail mask handling
- Use static array for generating tail masks (as requested)
- Apply tail mask to the weights as well as the input vectors to prevent spurious propagation of NaNs/Infs
* Revert whitespace change
* Improve readability of conditions for using AVX
* dnn(lstm): minor coding style changes, replaced left aligned load
Support non-zero hidden state for LSTM
* fully support non-zero hidden state for LSTM
* check dims of hidden state for LSTM
* fix failed test Test_Model.TextRecognition
* add new tests for LSTM w/ non-zero hidden params
Co-authored-by: Julie Bareeva <julia.bareeva@xperience.ai>
* Remove a forward method in dnn::Layer
* Add a test
* Fix tests
* Mark multiple dnn::Layer::finalize methods as deprecated
* Replace back dnn's inputBlobs to vector of pointers
* Remove Layer::forward_fallback from CV_OCL_RUN scopes
Add layer forward interface with InputArrayOfArrays and
OutputArrayOfArrays parameters, it allows UMat buffer to be
processed and transferred in the layers.
Signed-off-by: Li Peng <peng.li@intel.com>
* another round of dnn optimization:
* increased malloc alignment across OpenCV from 16 to 64 bytes to make it AVX2 and even AVX-512 friendly
* improved SIMD optimization of pooling layer, optimized average pooling
* cleaned up convolution layer implementation
* made activation layer "attacheable" to all other layers, including fully connected and addition layer.
* fixed bug in the fusion algorithm: "LayerData::consumers" should not be cleared, because it desctibes the topology.
* greatly optimized permutation layer, which improved SSD performance
* parallelized element-wise binary/ternary/... ops (sum, prod, max)
* also, added missing copyrights to many of the layer implementation files
* temporarily disabled (again) the check for intermediate blobs consistency; fixed warnings from various builders
rogday
Alexander Alekhin
Andrew Ryrie
Julia Bareeva
Dmitry Kurtaev
Andrew Ryrie
Alexander Alekhin
Maksim Shabunin
Li Peng
Vadim Pisarevsky