The onnx model was loadable in 4.5.4 but not in 4.6.0

[iHaibaraAi]

opencv 4.5.4 and 4.6.0
Operating System / Platform => Windows 11 64 bit
Compiler => Visual Studio 2022

onnx
https://github.com/afterdm/onnx/blob/master/yunet.onnx

[ INFO:0@20.685] global C:\build\opencv-4.6.0\modules\dnn\src\onnx\onnx_importer.cpp (797) cv::dnn::dnn4_v20220524::ONNXImporter::populateNet DNN/ONNX: loading ONNX v6 model produced by 'pytorch':1.9. Number of nodes = 111, initializers = 92, inputs = 1, outputs = 3
[ INFO:0@20.685] global C:\build\opencv-4.6.0\modules\dnn\src\onnx\onnx_importer.cpp (713) cv::dnn::dnn4_v20220524::ONNXImporter::parseOperatorSet DNN/ONNX: ONNX opset version = 9
[ERROR:0@20.708] global C:\build\opencv-4.6.0\modules\dnn\src\onnx\onnx_importer.cpp (2845) cv::dnn::dnn4_v20220524::ONNXImporter::parseShape DNN/ONNX(Shape): dynamic 'zero' shapes are not supported, input 243 [ 0 0 0 51 ]
OpenCV(4.6.0) Error: Assertion failed (!isDynamicShape) in cv::dnn::dnn4_v20220524::ONNXImporter::parseShape, file C:\build\opencv-4.6.0\modules\dnn\src\onnx\onnx_importer.cpp, line 2846
[ERROR:0@20.716] global C:\build\opencv-4.6.0\modules\dnn\src\onnx\onnx_importer.cpp (1021) cv::dnn::dnn4_v20220524::ONNXImporter::handleNode DNN/ONNX: ERROR during processing node with 1 inputs and 1 outputs: [Shape]:(onnx_node!Shape_70) from domain='ai.onnx'
[ INFO:0@20.716] global C:\build\opencv-4.6.0\modules\dnn\src\onnx\onnx_importer.cpp (1025) cv::dnn::dnn4_v20220524::ONNXImporter::handleNode Input[0] = '243'
[ INFO:0@20.716] global C:\build\opencv-4.6.0\modules\dnn\src\onnx\onnx_importer.cpp (1029) cv::dnn::dnn4_v20220524::ONNXImporter::handleNode Output[0] = '265'
OpenCV(4.6.0) Error: Unspecified error (> Node [Shape@ai.onnx]:(onnx_node!Shape_70) parse error: OpenCV(4.6.0) C:\build\opencv-4.6.0\modules\dnn\src\onnx\onnx_importer.cpp:2846: error: (-215:Assertion failed) !isDynamicShape in function 'cv::dnn::dnn4_v20220524::ONNXImporter::parseShape'

) in cv::dnn::dnn4_v20220524::ONNXImporter::handleNode, file C:\build\opencv-4.6.0\modules\dnn\src\onnx\onnx_importer.cpp, line 1040

[zihaomu]

Thanks @afterdm, this bug is caused by the shape layer. I'll try to fix it now.

[fengyuentau]

OpenCV DNN does not support ONNX models with dynamic input shape. However, you can load a ONNX model with fixed input shape and infer with other input shapes using OpenCV DNN. You can download YuNet with fixed size input and demos from https://github.com/fengyuentau/opencv_zoo/tree/master/models/face_detection_yunet.

[fisakhan]

@fengyuentau thanks. How can I convert a model to fixed size input?
I convert a variable size input model (RetinaFace) from SavedModel (.pb) format to fixed.onnx and by loading it using net = cv2.dnn.readNet("fixed.onnx") I get the following error:

[ERROR:0@0.051] global /opt/conda/conda-bld/opencv-suite_1659973592936/work/modules/dnn/src/onnx/onnx_importer.cpp (2516) parseShape DNN/ONNX(Shape): dynamic 'zero' shapes are not supported, input input_image [ 0 0 0 3 ]
[ERROR:0@0.051] global /opt/conda/conda-bld/opencv-suite_1659973592936/work/modules/dnn/src/onnx/onnx_importer.cpp (906) handleNode DNN/ONNX: ERROR during processing node with 2 inputs and 1 outputs: [Squeeze]:(StatefulPartitionedCall/RetinaFaceModel/tf_op_layer_strided_slice_14/strided_slice_14__128:0) from domain='ai.onnx'

[fengyuentau]

[ERROR:0@0.051] global /opt/conda/conda-bld/opencv-suite_1659973592936/work/modules/dnn/src/onnx/onnx_importer.cpp (2516) parseShape DNN/ONNX(Shape): dynamic 'zero' shapes are not supported, input input_image [ 0 0 0 3 ]

@fisakhan This error is exactly caused by trying to load a dynamic shape ONNX model with OpenCV DNN. You can have a try with https://github.com/daquexian/onnx-simplifier for your purpose.

[fisakhan]

@fengyuentau After conversion using onnx-simplifier, I still get the same error. You can find both the models (savedmodel and onnx) at https://github.com/fisakhan/Face_Pose/tree/master/models. netron.app shows this for .onnx model

[fengyuentau]

@fisakhan Obviously the input shape is still dynamic. You need to set it fixed using onnx-simplifier.

[fisakhan]

Now I receive this error:

[ERROR:0@114.883] global /opt/conda/conda-bld/opencv-suite_1659973592936/work/modules/dnn/src/onnx/onnx_importer.cpp (906) handleNode DNN/ONNX: ERROR during processing node with 3 inputs and 1 outputs: [Clip]:(StatefulPartitionedCall/RetinaFaceModel/MobileNetV2_extrator/Conv1_relu/Relu6:0) from domain='ai.onnx'
Traceback (most recent call last):
  File "/media/Disk/Projects/opencv/scr/image.py", line 147, in <module>
    net = cv2.dnn.readNet("/media/Disk/model_conversion/models/retinaface_mbv2_sim.onnx")

error: OpenCV(4.5.5) /opt/conda/conda-bld/opencv-suite_1659973592936/work/modules/dnn/src/onnx/onnx_importer.cpp:928: error: (-2:Unspecified error) in function 'handleNode'
> Node [Clip@ai.onnx]:(StatefulPartitionedCall/RetinaFaceModel/MobileNetV2_extrator/Conv1_relu/Relu6:0) parse error: OpenCV(4.5.5) /opt/conda/conda-bld/opencv-suite_1659973592936/work/modules/dnn/src/onnx/onnx_importer.cpp:1613: error: (-2:Unspecified error) in function 'void cv::dnn::dnn4_v20211220::ONNXImporter::parseClip(cv::dnn::dnn4_v20211220::LayerParams&, const opencv_onnx::NodeProto&)'
> >  (expected: 'node_proto.input_size() == 1'), where
> >     'node_proto.input_size()' is 3
> > must be equal to
> >     '1' is 1
> 

[fengyuentau]

@fisakhan Please have a try with the latest OpenCV (for python: pip install opencv-python-rolling). This issue should have been resolved in #22100.

[fisakhan]

Now I get the following error:

model = cv2.dnn.readNetFromONNX('./models/rf640_simplified.onnx')

error: OpenCV(4.6.0) /io/opencv/modules/dnn/src/onnx/onnx_importer.cpp:1040: error: (-2:Unspecified error) in function 'handleNode'

Node [Shape@ai.onnx]:(onnx_node!StatefulPartitionedCall/RetinaFaceModel/tf_op_layer_Shape_2/Shape_2) parse error: OpenCV(4.6.0) /io/opencv/modules/dnn/src/onnx/onnx_importer.cpp:2846: error: (-215:Assertion failed) !isDynamicShape in function 'parseShape'