Change an order of PriorBox layer operations.

Fix batch norm in training phase.

during opencv_world builds

Signed-off-by: Wu, Zhiwen <zhiwen.wu@intel.com>
Signed-off-by: Li Peng <peng.li@intel.com>

Signed-off-by: Li Peng <peng.li@intel.com>

* ave pooling ocl fix

support the padded area control in ave pooling
Signed-off-by: Li Peng <peng.li@intel.com>

* warning fix: ununitialized field

Signed-off-by: Li Peng <peng.li@intel.com>

It fixes tensorflow ocl testcase of MobileNetSSD and Inception_v2_SSD
Signed-off-by: Li Peng <peng.li@intel.com>

include relu6 ocl kernel and layer fusion support
Signed-off-by: Li Peng <peng.li@intel.com>

* Use only absolute prior boxes explicit sizes. Remove scales attributes.

* Simplified PriorBox layer forward pass

The ocl convolution doesn't support tensorflow padMode well.
Add fallback check if we meet this situation, it could fix the
tensorflow MobileNet SSD failure.
Signed-off-by: Li Peng <peng.li@intel.com>

Signed-off-by: Li Peng <peng.li@intel.com>

* fix faster_rcnn sample crashed at PoolingInvoker operator() of pooling_layer.

* find_odj onmouse bug about find matched point status.

* reverted AutoBuffer back to std::vector

Found via `codespell`

Signed-off-by: Li Peng <peng.li@intel.com>

* Intel Inference Engine deep learning backend.

* OpenFace network using Inference Engine backend

it fuse ocl kernels to reduce kernel enqueue
Signed-off-by: Li Peng <peng.li@intel.com>

Signed-off-by: Li Peng <peng.li@intel.com>

Signed-off-by: Li Peng <peng.li@intel.com>

* Add a 512 bit codepath to the AVX512 fastConv function

this patch adds a 512 wide codepath to the fastConv() function for
AVX512 use.
The basic idea is to process the first N * 16 elements of the vector
with avx512, and then run the rest of the vector using the traditional
AVX2 codepath.

* dnn: use unaligned AVX512 load (OpenCV aligns data on 32-byte boundary)

* dnn: change "vecsize" condition for AVX512

* dnn: fix indentation

Signed-off-by: Li Peng <peng.li@intel.com>