Parsing ONNX with AND layer fails. #24979
Description
System Information
OpenCV python version: 5.x
Operating System / Mac OS Sonoma 14.3
Python version: 3.8.18
Detailed description
Parsing ONNX graph with AND layer fails. Error log:
[ INFO:0@0.545] global onnx_importer.cpp:806 populateNet DNN/ONNX: loading ONNX v8 model produced by 'pytorch':2.2.0. Number of nodes = 1, initializers = 0, inputs = 2, outputs = 1
[ INFO:0@0.545] global onnx_importer.cpp:699 parseOperatorSet DNN/ONNX: ONNX opset version = 17
[ INFO:0@0.548] global onnx_importer.cpp:977 handleNode DNN/ONNX: processing node with 2 inputs and 1 outputs: [And]:(onnx_node!/And) from domain='ai.onnx'
[ INFO:0@0.548] global onnx_importer.cpp:3307 parseCustomLayer DNN/ONNX: unknown node type, try using custom handler for node with 2 inputs and 1 outputs: [And]:(onnx_node!/And)
OpenCV(5.0.0-pre) Error: Unspecified error (Can't create layer "onnx_node!/And" of type "And") in getLayerInstance, file /Users/abd/work/opencv/modules/dnn/src/net_impl.hpp, line 106
[ERROR:0@0.549] global onnx_importer.cpp:1035 handleNode DNN/ONNX: ERROR during processing node with 2 inputs and 1 outputs: [And]:(onnx_node!/And) from domain='ai.onnx'
[ INFO:0@0.549] global onnx_importer.cpp:1039 handleNode Input[0] = 'input_x'
[ INFO:0@0.549] global onnx_importer.cpp:1039 handleNode Input[1] = 'input_y'
[ INFO:0@0.549] global onnx_importer.cpp:1043 handleNode Output[0] = 'output'
OpenCV(5.0.0-pre) Error: Unspecified error (> Node [And@ai.onnx]:(onnx_node!/And) parse error: OpenCV(5.0.0-pre) /Users/abd/work/opencv/modules/dnn/src/net_impl.hpp:106: error: (-2:Unspecified error) Can't create layer "onnx_node!/And" of type "And" in function 'getLayerInstance'
> ) in handleNode, file /Users/abd/work/opencv/modules/dnn/src/onnx/onnx_importer.cpp, line 1054
Traceback (most recent call last):
File "genOnnx.py", line 50, in <module>
net = cv2.dnn.readNetFromONNX(f"{name}.onnx");
cv2.error: OpenCV(5.0.0-pre) /Users/abd/work/opencv/modules/dnn/src/onnx/onnx_importer.cpp:1054: error: (-2:Unspecified error) in function 'handleNode'
> Node [And@ai.onnx]:(onnx_node!/And) parse error: OpenCV(5.0.0-pre) /Users/abd/work/opencv/modules/dnn/src/net_impl.hpp:106: error: (-2:Unspecified error) Can't create layer "onnx_node!/And" of type "And" in function 'getLayerInstance'
Steps to reproduce
import cv2
import torch
import torch.nn as nn
import numpy as np
class And(nn.Module):
def __init__(self):
super(And, self).__init__()
def forward(self, x, y):
return x & y
if __name__ == '__main__':
name = "and"
model = And()
x = torch.ones(5, dtype=torch.bool )
y = torch.zeros(5, dtype=torch.bool )
inputs = [x, y]
outputs = [model(x, y)]
print("\n\t\t **** TORCH INFERENCE END ****\n\n")
torch.onnx.export(
model,
(x, y),
f"{name}.onnx",
verbose=True,
input_names=[f"input_{i}" for i in range(len(inputs))],
output_names=["output"]
)
print("\n\t\t **** EXPORTED ONNX ****\n\n")
## save the inputs and outpus as npy files
for i in range(len(inputs)):
inputs[i] = inputs[i].detach().numpy()
with open(f"input_{name}_{i}.npy", "wb") as f:
np.save(f, inputs[i])
for i in range(len(outputs)):
outputs[i] = outputs[i].detach().numpy()
with open(f"output_{name}_{i}.npy", "wb") as f:
np.save(f, outputs[i])
for i in range(len(inputs)):
inputs[i] = cv2.Mat(inputs[i])
net = cv2.dnn.readNetFromONNX(f"{name}.onnx");
print("\n\t\t **** CV2 MODEL PARSED ****\n\n")
## save the inputs and outpus as npy files
for i in range(len(inputs)):
inputs[i] = inputs[i].detach().numpy()
with open(f"input_{name}_{i}.npy", "wb") as f:
np.save(f, inputs[i])
for i in range(len(outputs)):
outputs[i] = outputs[i].detach().numpy()
with open(f"output_{name}_{i}.npy", "wb") as f:
np.save(f, outputs[i])
for i in range(len(inputs)):
inputs[i] = cv2.Mat(inputs[i])
## set inputs
for i in range(len(inputs)):
net.setInput(inputs[i], f"input_{name}_{i}")
print("\n\t\t **** CV2 INPUTS SET ****\n\n")
## forward
output_cv = net.forward()
print("\n\t\t **** CV2 FORWARD END ****\n\n")
if len(outputs) == 1:
print(np.allclose(output_cv[0], outputs[0], atol=1e-6))
else:
for i in range(len(outputs)):
print(np.allclose(output_cv[i], outputs[i], atol=1e-6))
## print in green test PASSED
print("\033[92m\t\t **** TEST PASSED ****\n\n")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)