Awesome

[WIP] components_of_onnx

https://github.com/onnx/onnx/blob/main/docs/Operators.md

https://github.com/PINTO0309/simple-onnx-processing-tools

Key concept

• Convert each and every operator in Operators.md into a single ONNX file and commit.
• Use sog4onnx to generate the ONNX model for a single OP.
• Use sam4onnx to rewrite variables and constants.
• By preparing a large number of ONNX files of a single OP as a component and rewriting them individually with sam4onnx to the necessary parameters, any combination of OPs can be generated like a puzzle.
• All generated single-OP ONNX files are combined with snc4onnx.
• Use sna4onnx to add OPs to the generated ONNX graph and snd4onnx to remove OPs from the graph.
• The real goal of this repository is to componentize routine pre-processing and post-processing and directly replace the corresponding processing in various types of ONNX models, allowing for flexible switching between operations without modifying PyTorch or TensorFlow logic or re-exporting ONNX.

Base

As of April 28, 2022

• https://github.com/onnx/onnx/blob/fb1f72159f4030ce45503279e6f6b8a7fb103e0f/docs/Operators.md#aionnx-default

Opsets List

• onnx_opsets
• onnx_opsets.csv
• onnx_opsets.json
• opsets csv to json converter

Progress

<details><summary>Progress List</summary><div>

No.	Operator	Structure	Fin
001	Abs		:heavy_check_mark:
002	Acos		:heavy_check_mark:
003	Acosh		:heavy_check_mark:
004	Add		:heavy_check_mark:
005	And		:heavy_check_mark:
006	ArgMax		:heavy_check_mark:
007	ArgMin		:heavy_check_mark:
008	Asin		:heavy_check_mark:
009	Asinh		:heavy_check_mark:
010	Atan		:heavy_check_mark:
011	Atanh		:heavy_check_mark:
012	AveragePool
013	BatchNormalization
014	BitShift
015	Cast
016	Ceil
017	Clip
018	Compress
019	Concat
020	ConcatFromSequence
021	Constant
022	ConstantOfShape
023	Conv
024	ConvInteger
025	ConvTranspose
026	Cos
027	Cosh
028	CumSum
029	DepthToSpace
030	DequantizeLinear
031	Det
032	Div
033	Dropout
034	Einsum
035	Elu
036	Equal
037	Erf
038	Exp
039	Expand
040	EyeLike
041	Flatten
042	Floor
043	GRU
044	Gather
045	GatherElements
046	GatherND
047	Gemm
048	GlobalAveragePool
049	GlobalLpPool
050	GlobalMaxPool
051	Greater
052	GridSample
053	HardSigmoid
054	Hardmax
055	Identity
056	If
057	InstanceNormalization
058	IsInf
059	IsNaN
060	LRN
061	LSTM
062	LeakyRelu
063	Less
064	Log
065	Loop
066	LpNormalization
067	LpPool
068	MatMul
069	MatMulInteger
070	Max
071	MaxPool
072	MaxRoiPool
073	MaxUnpool
074	Mean
075	Min
076	Mod
077	Mul
078	Multinomial
079	Neg
080	NonMaxSuppression		:heavy_check_mark:
081	NonZero
082	Not
083	OneHot
084	Optional
085	OptionalGetElement
086	OptionalHasElement
087	Or
088	PRelu
089	Pad
090	Pow
091	QLinearConv
092	QLinearMatMul
093	QuantizeLinear
094	RNN
095	RandomNormal
096	RandomNormalLike
097	RandomUniform
098	RandomUniformLike
099	Reciprocal
100	ReduceL1
101	ReduceL2
102	ReduceLogSum
103	ReduceLogSumExp
104	ReduceMax
105	ReduceMean
106	ReduceMin
107	ReduceProd
108	ReduceSum
109	ReduceSumSquare
110	Relu
111	Reshape
112	Resize		:heavy_check_mark:
113	ReverseSequence
114	RoiAlign
115	Round
116	Scan
117	Scatter
118	ScatterElements
119	ScatterND
120	Selu
121	SequenceAt
122	SequenceConstruct
123	SequenceEmpty
124	SequenceErase
125	SequenceInsert
126	SequenceLength
127	Shape
128	Shrink
129	Sigmoid
130	Sign
131	Sin
132	Sinh
133	Size
134	Slice
135	Softplus
136	Softsign
137	SpaceToDepth
138	Split
139	SplitToSequence
140	Sqrt
141	Squeeze
142	StringNormalizer
143	Sub
144	Sum
145	Tan
146	Tanh
147	TfIdfVectorizer
148	ThresholdedRelu
149	Tile
150	TopK
151	Transpose
152	Trilu
153	Unique
154	Unsqueeze
155	Upsample
156	Where
157	Xor
158	Bernoulli
159	CastLike
160	Celu
161	DynamicQuantizeLinear
162	GreaterOrEqual
163	HardSwish
164	LessOrEqual
165	LogSoftmax
166	MeanVarianceNormalization
167	NegativeLogLikelihoodLoss
168	Range
169	SequenceMap
170	Softmax
171	SoftmaxCrossEntropyLoss
Z001	Normalization_rgb_imagenet		:heavy_check_mark:
Z002	Normalization_bgr_imagenet		:heavy_check_mark:
Z003	SingleClass_NonMaxSupression	480x640	:heavy_check_mark:
Z004	YOLACT_Edge_NonMaxSupression	550x550<br>	:heavy_check_mark:
Z005	Resize_0.5x0.5	1x3xHxW->1x3x(Hx0.5)x(Wx0.5)<br>	:heavy_check_mark:
Z006	Resize_HxW	1x3xHxW->1x3x(Hx?)x(Wx?)<br>	:heavy_check_mark:
Z007	Myriad_workaround_NonMaxSuppression		:heavy_check_mark:
Z008	TensorRT_compatible_N_batch_Resize		:heavy_check_mark:
Z009	Unity_Barracuda_compatible_GatherND		:heavy_check_mark:
Z010	Unity_Barracuda_compatible_Split		:heavy_check_mark:
Z011	YOLACT_PostProcess		:heavy_check_mark:
Z012	DAMO-YOLO_PostProcess		:heavy_check_mark:
Z013	YOLO_General_PostProcess (anchor+NMS)		:heavy_check_mark:
Z014	Inverse (com.microsoft)		:heavy_check_mark:
Z015	MSELoss		:heavy_check_mark:
Z016	EfficientNMS_TRT		:heavy_check_mark:
Z017	BatchedNMS_TRT		:heavy_check_mark:
Z018	AffineTransform		:heavy_check_mark:
Z019	AffineGrid_4D (general purpose)		:heavy_check_mark:
Z020	WarpAffine_4D		:heavy_check_mark:

Reference

1. https://stackoverflow.com/questions/47344571/how-to-draw-checkbox-or-tick-mark-in-github-markdown-table
2. Samples of handcrafting ONNX models (base64 encoding, base64 decoding, etc.)