mirror of
https://github.com/opencv/opencv.git
synced 2026-07-31 00:03:03 +04:00
Merge pull request #21865 from rogday:nary_eltwise_layers
Reimplementation of Element-wise layers with broadcasting support * init * semi-working initial version * add small_vector * wip * remove smallvec * add nary function * replace auto with Mat in lambda expr used in transform * uncomment asserts * autobuffer shape_buf & step_buf * fix a missing bracket * fixed a missing addLayer in parseElementWise * solve one-dimensional broadcast * remove pre_broadcast_transform for the case of two constants; fix missing constBlobsExtraInfo when addConstant is called * one autobuffer for step & shape * temporal fix for the missing original dimension information * fix parseUnsqueeze when it gets a 1d tensor constant * support sum/mean/min/max with only one input * reuse old code to handle cases of two non-constant inputs * add condition to handle div & mul of two non-constant inputs * use || instead of or * remove trainling spaces * enlarge buf in binary_forward to contain other buffer * use autobuffer in nary_forward * generate data randomly and add more cases for perf * add op and, or & xor * update perf_dnn * remove some comments * remove legacy; add two ONNX conformance tests in filter * move from cpu_denylist to all_denylist * adjust parsing for inputs>=2 Co-authored-by: fengyuentau <yuantao.feng@opencv.org.cn>
This commit is contained in:
rogday
GitHub
@@ -55,7 +55,156 @@ struct Layer_Slice : public TestBaseWithParam<tuple<Backend, Target> >
|
||||
}
|
||||
};
|
||||
|
||||
struct Layer_NaryEltwise : public TestBaseWithParam<tuple<Backend, Target> >
|
||||
{
|
||||
void test_layer(const std::vector<int>& a_shape, const std::vector<int>& b_shape, const String op, bool isRef = false)
|
||||
{
|
||||
int backendId = get<0>(GetParam());
|
||||
int targetId = get<1>(GetParam());
|
||||
|
||||
Mat a(a_shape, CV_32FC1);
|
||||
Mat b(b_shape, CV_32FC1);
|
||||
|
||||
Scalar mean = 0.f;
|
||||
Scalar std = 1.f;
|
||||
randn(a, mean, std);
|
||||
randn(b, mean, std);
|
||||
|
||||
|
||||
Net net;
|
||||
LayerParams lp;
|
||||
if (isRef)
|
||||
lp.type = "Eltwise";
|
||||
else
|
||||
lp.type = "NaryEltwise";
|
||||
lp.name = "testLayer";
|
||||
lp.set("operation", op);
|
||||
int id = net.addLayerToPrev(lp.name, lp.type, lp);
|
||||
net.connect(0, 1, id, 1);
|
||||
|
||||
// warmup
|
||||
{
|
||||
std::vector<String> inpNames(2);
|
||||
inpNames[0] = "a";
|
||||
inpNames[1] = "b";
|
||||
net.setInputsNames(inpNames);
|
||||
net.setInput(a, inpNames[0]);
|
||||
net.setInput(b, inpNames[1]);
|
||||
|
||||
net.setPreferableBackend(backendId);
|
||||
net.setPreferableTarget(targetId);
|
||||
Mat out = net.forward();
|
||||
}
|
||||
|
||||
TEST_CYCLE()
|
||||
{
|
||||
Mat res = net.forward();
|
||||
}
|
||||
|
||||
SANITY_CHECK_NOTHING();
|
||||
}
|
||||
|
||||
int N = 8;
|
||||
int C = 256;
|
||||
int H = 128;
|
||||
int W = 100;
|
||||
};
|
||||
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_add)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "add");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_div)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "div");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_ref_div)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "div", true);
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_equal)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "equal");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_greater)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "greater");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_less)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "less");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_max)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "max");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_ref_max)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "max", true);
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_mean)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "mean");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_min)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "min");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_ref_min)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "min", true);
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_mul)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "mul");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_ref_mul)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "prod", true);
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_pow)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "pow");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_sub)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "sub");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_sum)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "sum");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_NCHW_ref_sum)
|
||||
{
|
||||
test_layer({N, C, H, W}, {N, C, H, W}, "sum", true);
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NCHW_C_sum)
|
||||
{
|
||||
test_layer({N, C, H, W}, {C, 1, 1}, "sum");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_NaryEltwise, NHWC_C)
|
||||
{
|
||||
test_layer({N, H, W, C}, {1, C}, "sum");
|
||||
}
|
||||
|
||||
PERF_TEST_P_(Layer_Slice, YOLOv4_tiny_1)
|
||||
{
|
||||
@@ -91,5 +240,6 @@ PERF_TEST_P_(Layer_Slice, FastNeuralStyle_eccv16)
|
||||
}
|
||||
|
||||
INSTANTIATE_TEST_CASE_P(/**/, Layer_Slice, dnnBackendsAndTargets(false, false));
|
||||
INSTANTIATE_TEST_CASE_P(/**/, Layer_NaryEltwise, testing::Values(std::make_tuple(DNN_BACKEND_OPENCV, DNN_TARGET_CPU)));
|
||||
|
||||
} // namespace
|
||||
|
||||
Reference in New Issue
Block a user