ONNX Support Status

Note:In this document, the numbers in the header of all tables represent the version of onnx opset.

Import

  • ✓: onnx specification defined, and supported.
  • X: onnx specification defined, but not support yet.
  • Empty: Not defined (Support status follows latest).
ONNX Version Info:
 
  • Version: 1.4.1

Total: 81/129

ONNX Operator 1 2 3 4 5 6 7 8 9 Description
Abs                
Acos                  
Acosh                  
Add              
And              
ArgMax                
ArgMin                
Asin                  
Asinh                  
Atan                  
Atanh                  
AveragePool             Not all features are verified, since some features are not supported by caffe2. Those features can be verified by ONNXRuntime when opset > 6. Some feature is not supported by Nnabla such as Pad’s edge mode.
BatchNormalization X         X X    
Cast X         X     X  
Ceil                
Clip                
Compress                 X  
Concat              
Constant             X  
ConstantOfShape                 X  
Conv                
ConvTranspose                
Cos                  
Cosh                  
DepthToSpace                
Div              
Dropout X         X      
Elu                
Equal              
Erf                 X  
Exp                
Expand               X    
EyeLike                 X  
Flatten              
Floor                
GRU X   X       X      
Gather X                  
Gemm            
GlobalAveragePool                
GlobalLpPool X X                
GlobalMaxPool X                  
Greater            
HardSigmoid X         X        
Hardmax X                  
Identity                
If X                  
InstanceNormalization X         X        
IsNaN                  
LRN                
LSTM X           X      
LeakyRelu                
Less            
Log                
LogSoftmax                
Loop X                  
LpNormalization X                  
LpPool X X                
MatMul              
Max            
MaxPool           X   Not all features are verified, since some features are not supported by caffe2. Those features can be verified by ONNXRuntime.
MaxRoiPool X                  
MaxUnpool                 X  
Mean            
Min            
Mul              
Multinomial             X      
Neg                
NonZero                 X  
Not                
OneHot                 X  
Or              
PRelu         X   X  
Pad             Onnx required to support “edge” mode, while nnabla does not support it.
Pow              
RNN X           X      
RandomNormal X                  
RandomNormalLike X                  
RandomUniform X                  
RandomUniformLike X                  
Reciprocal                
ReduceL1 X                  
ReduceL2 X                  
ReduceLogSum X                  
ReduceLogSumExp X                  
ReduceMax                
ReduceMean                
ReduceMin                
ReduceProd                
ReduceSum                
ReduceSumSquare X                  
Relu                
Reshape              
Resize                    
Scan               X X  
Scatter                 X  
Selu                
Shape X                  
Shrink                 X  
Sigmoid                
Sign                  
Sin                  
Sinh                  
Size X                  
Slice                
Softmax                
Softplus                
Softsign                
SpaceToDepth                
Split              
Sqrt                
Squeeze                
StringNormalizer                    
Sub              
Sum            
Tan                  
Tanh                
TfIdfVectorizer                 X  
ThresholdedRelu                    
Tile                
TopK X                  
Transpose                
Unsqueeze                
Upsample            
Where                 X  
Xor              
experimental ATen X                  
experimental GRUUnit X                  
experimental GivenTensorFill X                  
experimental Scale X                  

Export

  • ✓: Support to export this opset.
  • △: Partially support to export this opset (e.g. some cases cannot be supported, or not completely tested).
  • X: Supported, but test failed.
  • Empty: Not support corresponding opset version.
NNabla Version Info:
 
  • Version: 1.0.21

Total: 82/172

Count 10/14

NNabla Function 6 7 9 Description
Affine  
RNN        
LSTM        
GRU        
Convolution  
DepthwiseConvolution  
Deconvolution Caffe2 and onnxruntime do not support dilations != 1.
DepthwiseDeconvolution Caffe2 and onnxruntime do not support dilations != 1.
MaxPooling  
AveragePooling Currently only supports the cases where both ignore_border and including_pad are True.
GlobalAveragePooling  
SumPooling X  
Unpooling The kernel only supports 2d on opset 6.
Embed        

Count 8/21

NNabla Function 6 7 9 Description
Sigmoid  
Swish        
Tanh  
ReLU  
LeakyReLU  
Softmax ONNX Add, Sub operator does not support multidirectional broadcasting on opset 6.
LogSoftmax        
ELU  
SELU  
CReLU        
CELU        
PReLU  
GELU        
ReLU6        
HardSigmoid        
HardTanh        
LogSigmoid        
SoftPlus        
SoftSign        
TanhShrink        
Sinc        

Count 1/6

NNabla Function 6 7 9 Description
FusedBatchNormalization        
BatchNormalization In inferring stage, caffe2 mistmatch onnx 1.4.x’s implementation, “in-place” feature cannot be applied.
SyncBatchNormalization        
MeanSubtraction        
ClipGradByValue        
ClipGradByNorm        

Count 5/7

NNabla Function 6 7 9 Description
Sum  
Mean  
Max  
Min  
Prod  
ReduceSum        
ReduceMean        

Count 11/12

NNabla Function 6 7 9 Description
Add2 ONNX Add operator does not support multidirectional broadcasting on opset 6.
BcAdd2        
Sub2 ONNX Sub operator does not support multidirectional broadcasting on opset 6.
Mul2 ONNX Mul operator does not support multidirectional broadcasting on opset 6.
Div2 ONNX Div operator does not support multidirectional broadcasting on opset 6.
Pow2 ONNX Pow operator does not support multidirectional broadcasting on opset 6.
AddScalar  
MulScalar  
PowScalar  
RSubScalar  
RDivScalar  
RPowScalar  

Count 12/29

NNabla Function 6 7 9 Description
Sign X X  
Minimum2 ONNX Add operator does not support multidirectional broadcasting on opset 6.
Maximum2 ONNX Add operator does not support multidirectional broadcasting on opset 6.
MinimumScalar  
MaximumScalar  
LogicalAnd  
LogicalOr  
LogicalXor  
Equal  
NotEqual        
GreaterEqual        
Greater  
LessEqual        
Less  
LogicalAndScalar        
LogicalOrScalar        
LogicalXorScalar        
EqualScalar        
NotEqualScalar        
GreaterEqualScalar        
GreaterScalar        
LessEqualScalar        
LessScalar        
LogicalNot  
IsNaN        
IsInf        
ResetNaN        
ResetInf        
Where        

Count 19/22

NNabla Function 6 7 9 Description
Constant        
Arange        
Abs  
Exp  
Log  
Identity  
BatchMatmul  
Round        
Ceil  
Floor  
Sin X  
Cos X  
Tan X  
Sinh X X  
Cosh X X  
ASin X  
ACos X  
ATan X  
ATan2 X  
ASinh X X  
ACosh X X  
ATanh X X  

Count 9/19

NNabla Function 6 7 9 Description
Concatenate  
Split  
Stack  
Slice ONNX slice cannot support step != 1 on opset < 10.
Pad When the mode of the pad is reflect, if the size of the pad exceeds the input size, caffe2 and onnxruntime cannot handle it.
Transpose  
Broadcast        
BroadcastTo  
Tile        
OneHot        
Flip  
Shift        
Sort        
Reshape  
MatrixDiag        
MatrixDiagPart        
Assign        
GatherNd        
ScatterNd        

Count 0/3

NNabla Function 6 7 9 Description
Interpolate        
FFT        
IFFT        

Count 1/11

NNabla Function 6 7 9 Description
Dropout  
TopKData        
TopKGrad        
Rand        
Randint        
Randn        
RandomChoice        
RandomCrop        
RandomFlip        
RandomShift        
ImageAugmentation        

Count 0/9

NNabla Function 6 7 9 Description
SigmoidCrossEntropy        
BinaryCrossEntropy        
SoftmaxCrossEntropy        
CategoricalCrossEntropy        
SquaredError        
AbsoluteError        
HuberLoss        
EpsilonInsensitiveLoss        
KLMultinomial        

Count 6/11

NNabla Function 6 7 9 Description
BinarySigmoid X X  
BinaryTanh X X  
BinaryConnectAffine  
BinaryConnectConvolution  
BinaryWeightAffine  
BinaryWeightConvolution  
INQAffine        
INQConvolution        
FixedPointQuantize        
Pow2Quantize        
Prune        

Count 0/3

NNabla Function 6 7 9 Description
TopNError        
BinaryError        
ConfusionMatrix        

Count 0/5

NNabla Function 6 7 9 Description
VATNoise        
Unlink        
Sink        
NmsDetection2d        
MaxPoolingBackward