dnn: inference results are different from ONNX Runtime on the InstanceNorm layer #24377
Description
System Information
latest opencv
Detailed description
Relates #24092 (comment)
TEST_P(Test_ONNX_layers, InstanceNormEpsilon) {
testONNXModels("test_instancenorm_epsilon", pb);
}
TEST_P(Test_ONNX_layers, InstanceNormExample) {
testONNXModels("test_instancenorm_example", pb);
}$ ./build/bin/opencv_test_dnn --gtest_filter="Test_ONNX_layers.InstanceNorm*"
CTEST_FULL_OUTPUT
OpenCV version: 4.8.0-dev
OpenCV VCS version: 4.8.0-299-g590f150d5e
Build type: Release
Compiler: /usr/bin/c++ (ver 9.4.0)
Parallel framework: pthreads (nthreads=6)
CPU features: SSE SSE2 SSE3 *SSE4.1 *SSE4.2 *FP16? *AVX *AVX2 *AVX512-SKX?
Intel(R) IPP version: ippIP SSE4.2 (y8) 2021.8 (-) Feb 20 2023
Intel(R) IPP features code: 0x80
OpenCL is disabled
TEST: Skip tests with tags: 'mem_6gb', 'verylong', 'dnn_skip_opencv_backend', 'dnn_skip_cpu', 'dnn_skip_cpu_fp16', 'dnn_skip_ocl', 'dnn_skip_ocl_fp16', 'dnn_skip_onnx_conformance', 'dnn_skip_parser'
Note: Google Test filter = Test_ONNX_layers.InstanceNorm*
[==========] Running 3 tests from 1 test case.
[----------] Global test environment set-up.
[----------] 3 tests from Test_ONNX_layers
[ RUN ] Test_ONNX_layers.InstanceNorm/0, where GetParam() = OCV/CPU
[ OK ] Test_ONNX_layers.InstanceNorm/0 (1 ms)
[ RUN ] Test_ONNX_layers.InstanceNormEpsilon/0, where GetParam() = OCV/CPU
/home/tao/workspace/fytao/workspace/instance_norm/opencv/modules/dnn/test/test_common.impl.hpp:76: Failure
Expected: (normL1) <= (l1), actual: 0.632857 vs 1e-05
test_instancenorm_epsilon |ref| = 2.7365410327911377
/home/tao/workspace/fytao/workspace/instance_norm/opencv/modules/dnn/test/test_common.impl.hpp:79: Failure
Expected: (normInf) <= (lInf), actual: 4.001 vs 0.0001
test_instancenorm_epsilon |ref| = 2.7365410327911377
[ FAILED ] Test_ONNX_layers.InstanceNormEpsilon/0, where GetParam() = OCV/CPU (0 ms)
[ RUN ] Test_ONNX_layers.InstanceNormExample/0, where GetParam() = OCV/CPU
[ERROR:0@0.018] global onnx_importer.cpp:1029 handleNode DNN/ONNX: ERROR during processing node with 3 inputs and 1 outputs: [InstanceNormalization]:(onnx_node_output_0!y) from domain='ai.onnx'
unknown file: Failure
C++ exception with description "OpenCV(4.8.0-dev) /home/tao/workspace/fytao/workspace/instance_norm/opencv/modules/dnn/src/onnx/onnx_importer.cpp:1051: error: (-2:Unspecified error) in function 'handleNode'
> Node [InstanceNormalization@ai.onnx]:(onnx_node_output_0!y) parse error: OpenCV(4.8.0-dev) /home/tao/workspace/fytao/workspace/instance_norm/opencv/modules/dnn/include/opencv2/dnn/dnn.inl.hpp:350: error: (-204:Requested object was not found) Required argument "epsilon" not found into dictionary in function 'get'
> " thrown in the test body.
[ FAILED ] Test_ONNX_layers.InstanceNormExample/0, where GetParam() = OCV/CPU (1 ms)
[----------] 3 tests from Test_ONNX_layers (2 ms total)
[----------] Global test environment tear-down
[==========] 3 tests from 1 test case ran. (2 ms total)
[ PASSED ] 1 test.
[ FAILED ] 2 tests, listed below:
[ FAILED ] Test_ONNX_layers.InstanceNormEpsilon/0, where GetParam() = OCV/CPU
[ FAILED ] Test_ONNX_layers.InstanceNormExample/0, where GetParam() = OCV/CPU
2 FAILED TESTSTest models and data are collected from ONNX conformance tests. They are converted to the form that opencv dnn tests use.
Steps to reproduce
- Prepare opencv with the above test code, build target
opencv_test_dnn. - Prepare opencv_extra with all.zip.
- Set environment and run the test with
./build/bin/opencv_test_dnn --gtest_filter="Test_ONNX_layers.InstanceNorm*".
Issue submission checklist
- I report the issue, it's not a question
- I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
- I updated to the latest OpenCV version and the issue is still there
- There is reproducer code and related data files (videos, images, onnx, etc)