optimize out scaleLayer & concatLayer whenever possible
authorVadim Pisarevsky <vadim.pisarevsky@gmail.com>
Tue, 4 Jul 2017 14:23:47 +0000 (17:23 +0300)
committerVadim Pisarevsky <vadim.pisarevsky@gmail.com>
Fri, 14 Jul 2017 15:30:53 +0000 (18:30 +0300)
commit0488d9bdb24b3b9f75003971e6b91eed2bcb474e
tree83cf4e607c4264ba13fb0b500dea248baec8e288
parent431e2e6d6895c84a5309bcce2b2939bd2bb443e8
optimize out scaleLayer & concatLayer whenever possible

fixed problem in concat layer by disabling memory re-use in layers with multiple inputs

trying to fix the tests when Halide is used to run deep nets

another attempt to fix Halide tests

see if the Halide tests will pass with concat layer fusion turned off

trying to fix failures in halide tests; another try

one more experiment to make halide_concat & halide_enet tests pass

continue attempts to fix halide tests

moving on

uncomment parallel concat layer

seemingly fixed failures in Halide tests and re-enabled concat layer fusion; thanks to dkurt for the patch
modules/dnn/include/opencv2/dnn/dnn.hpp
modules/dnn/src/dnn.cpp
modules/dnn/src/layers/concat_layer.cpp
modules/dnn/src/layers/convolution_layer.cpp
modules/dnn/test/test_halide_layers.cpp