mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-28 05:06:29 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #24069 from dkurt:openvino_detection_layer

Browse Source 
DetectionOutput layer on OpenVINO without limitations #24069

### Pull Request Readiness Checklist

required for https://github.com/opencv/opencv/pull/23987

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
This commit is contained in:
Dmitry Kurtaev 2023-08-02 14:28:47 +03:00 committed by GitHub
parent f46f7eff0c
commit 195aad8e6a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 58 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
modules/dnn/src/layers/detection_output_layer.cpp
View File

@ -221,7 +221,7 @@ public:
{
return backendId == DNN_BACKEND_OPENCV ||
(backendId == DNN_BACKEND_CUDA && !_groupByClasses) ||
(backendId == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && !_locPredTransposed && _bboxesNormalized);
backendId == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH;
}
bool getMemoryShapes(const std::vector<MatShape> &inputs,
@ -1006,9 +1006,30 @@ public:
virtual Ptr<BackendNode> initNgraph(const std::vector<Ptr<BackendWrapper> >& inputs, const std::vector<Ptr<BackendNode> >& nodes) CV_OVERRIDE
{
CV_Assert(nodes.size() == 3);
auto& box_logits = nodes[0].dynamicCast<InfEngineNgraphNode>()->node;
auto& class_preds = nodes[1].dynamicCast<InfEngineNgraphNode>()->node;
auto& proposals = nodes[2].dynamicCast<InfEngineNgraphNode>()->node;
auto box_logits = nodes[0].dynamicCast<InfEngineNgraphNode>()->node;
auto class_preds = nodes[1].dynamicCast<InfEngineNgraphNode>()->node;
auto proposals = nodes[2].dynamicCast<InfEngineNgraphNode>()->node;
if (_locPredTransposed) {
// Convert box predictions from yxYX to xyXY
box_logits = std::make_shared<ngraph::op::v1::Reshape>(box_logits,
std::make_shared<ngraph::op::Constant>(ngraph::element::i32, ngraph::Shape{3}, std::vector<int32_t>{0, -1, 2}),
true
);
int axis = 2;
box_logits = std::make_shared<ngraph::op::v1::Reverse>(box_logits,
std::make_shared<ngraph::op::Constant>(ngraph::element::i32, ngraph::Shape{1}, &axis),
ngraph::op::v1::Reverse::Mode::INDEX
);
}
auto shape = std::make_shared<ngraph::op::Constant>(ngraph::element::i32, ngraph::Shape{2}, std::vector<int32_t>{0, -1});
box_logits = std::make_shared<ngraph::op::v1::Reshape>(box_logits, shape, true);
class_preds = std::make_shared<ngraph::op::v1::Reshape>(class_preds, shape, true);
proposals = std::make_shared<ngraph::op::v1::Reshape>(proposals,
std::make_shared<ngraph::op::Constant>(ngraph::element::i32, ngraph::Shape{3}, std::vector<int32_t>{0, _varianceEncodedInTarget ? 1 : 2, -1}),
true
);
ngraph::op::DetectionOutputAttrs attrs;
attrs.num_classes = _numClasses;

29
modules/dnn/test/test_caffe_importer.cpp
View File

@ -731,7 +731,7 @@ TEST_P(Test_Caffe_nets, FasterRCNN_vgg16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
double scoreDiff = 0.0;
double scoreDiff = 0.0, iouDiff = 0.0;
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2022010000)
// Check 'backward_compatible_check || in_out_elements_equal' failed at core/src/op/reshape.cpp:427:
// While validating node 'v1::Reshape bbox_pred_reshape (bbox_pred[0]:f32{1,84}, Constant_265242[0]:i64{4}) -> (f32{?,?,?,?})' with friendly_name 'bbox_pred_reshape':
@ -741,11 +741,20 @@ TEST_P(Test_Caffe_nets, FasterRCNN_vgg16)
if (target == DNN_TARGET_OPENCL_FP16)
scoreDiff = 0.02;
#endif
#if defined(INF_ENGINE_RELEASE)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) {
iouDiff = 0.02;
if (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16) {
scoreDiff = 0.04;
iouDiff = 0.06;
}
}
#endif
static Mat ref = (Mat_<float>(3, 7) << 0, 2, 0.949398, 99.2454, 210.141, 601.205, 462.849,
0, 7, 0.997022, 481.841, 92.3218, 722.685, 175.953,
0, 12, 0.993028, 133.221, 189.377, 350.994, 563.166);
testFaster("faster_rcnn_vgg16.prototxt", "VGG16_faster_rcnn_final.caffemodel", ref, scoreDiff);
testFaster("faster_rcnn_vgg16.prototxt", "VGG16_faster_rcnn_final.caffemodel", ref, scoreDiff, iouDiff);
}
TEST_P(Test_Caffe_nets, FasterRCNN_zf)
@ -766,9 +775,6 @@ TEST_P(Test_Caffe_nets, FasterRCNN_zf)
);
#endif
if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ||
backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) && target == DNN_TARGET_OPENCL_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16);
if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ||
backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD);
@ -779,7 +785,14 @@ TEST_P(Test_Caffe_nets, FasterRCNN_zf)
static Mat ref = (Mat_<float>(3, 7) << 0, 2, 0.90121, 120.407, 115.83, 570.586, 528.395,
0, 7, 0.988779, 469.849, 75.1756, 718.64, 186.762,
0, 12, 0.967198, 138.588, 206.843, 329.766, 553.176);
testFaster("faster_rcnn_zf.prototxt", "ZF_faster_rcnn_final.caffemodel", ref);
double scoreDiff = 0.0, iouDiff = 0.0;
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) {
scoreDiff = 0.02;
iouDiff = 0.13;
}
testFaster("faster_rcnn_zf.prototxt", "ZF_faster_rcnn_final.caffemodel", ref, scoreDiff, iouDiff);
}
TEST_P(Test_Caffe_nets, RFCN)
@ -802,8 +815,8 @@ TEST_P(Test_Caffe_nets, RFCN)
iouDiff = 0.12;
}
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2022010000)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
#if defined(INF_ENGINE_RELEASE)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
{
scoreDiff = 0.1f;
iouDiff = 0.2f;

11
modules/dnn/test/test_model.cpp
View File

@ -447,14 +447,17 @@ TEST_P(Test_Model, DetectionOutput)
{
if (backend == DNN_BACKEND_OPENCV)
scoreDiff = 4e-3;
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2022010000)
else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
scoreDiff = 4e-2;
#endif
else
scoreDiff = 2e-2;
iouDiff = 1.8e-1;
}
#if defined(INF_ENGINE_RELEASE)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
{
scoreDiff = 0.05;
iouDiff = 0.08;
}
#endif
testDetectModel(weights_file, config_file, img_path, refClassIds, refConfidences, refBoxes,
scoreDiff, iouDiff, confThreshold, nmsThreshold, size, mean);

5
modules/dnn/test/test_tf_importer.cpp
View File

@ -1816,6 +1816,11 @@ TEST_P(Test_TensorFlow_nets, Mask_RCNN)
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD || target == DNN_TARGET_CPU_FP16) ? 0.2 : 2e-5;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD || target == DNN_TARGET_CPU_FP16) ? 0.018 : default_lInf;
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
{
scoreDiff = std::max(scoreDiff, 0.02);
iouDiff = std::max(iouDiff, 0.009);
}
normAssertDetections(refDetections, outDetections, "", /*threshold for zero confidence*/1e-5, scoreDiff, iouDiff);
// Output size of masks is NxCxHxW where