5.3 KiB
支持的框架与算子
基准版本: 6.33.x (2024-06-24)
ONNX
opset 7–19 (ONNX 1.14.0)
Abs, Add, And, ArgMax, ArgMin, Atan, Atanh, AveragePool, BatchNormalization, BitwiseAnd, BitwiseOr, BitwiseXor, Cast, CastLike, Ceil, Celu, CenterCropPad, Clip, Col2Im, Concat, ConstantOfShape, Conv, ConvTranspose, Cos, Cumsum, DepthToSpace, DequantizeLinear, DFT, Div, Dropout, Einsum, Elu, Equal, Erf, Exp, Expand, Flatten, Floor, Gather, GatherElements, GatherND, Gelu, Gemm, GlobalAveragePool, GlobalMaxPool, Greater, GreaterOrEqual, GridSample, GroupNormalization, GRU, HammingWindow, HannWindow, HardSigmoid, HardSwish, InstanceNormalization, LayerNormalization, LeakyRelu, Less, LessOrEqual, Log, LogSoftmax, LRN, LSTM, MatMul, MaxPool, MaxRoiPool, MeanVarianceNormalization, Mish, Mod, Mul, Neg, OneHot, Or, Pad, Pow, PRelu, QLinearConv, QLinearMatMul, QuantizeLinear, Range, Reciprocal, ReduceL1, ReduceL2, ReduceLogSum, ReduceLogSumExp, ReduceMax, ReduceMean, ReduceMin, ReduceProd, ReduceSum, ReduceSumSquare, Relu, Reshape, Resize, ReverseSequence, Round, Rsqrt, ScatterElements, ScatterND, Selu, Shape, Sigmoid, Sign, Silu, Sin, Size, Slice, Softmax, Softplus, Softsign, SpaceToDepth, Split, Sqrt, Square, Squeeze, STFT, Sub, Tan, Tanh, Tile, TopK, Transpose, Unsqueeze, Upsample, Where, Xor
TensorFlow
2.15.0
tf.abs, tf.add, tf.add_n, tf.batch_to_space_nd, tf.cast, tf.clip_by_value, tf.concat, tf.depth_to_space, tf.div, tf.divide, tf.equal, tf.erf, tf.exp, tf.floor, tf.floordiv, tf.gather, tf.gather_nd, tf.greater, tf.greater_equal, tf.less, tf.less_equal, tf.logical_add, tf.matmul, tf.batch_matmul, tf.maximum, tf.minimum, tf.multiply, tf.negative, tf.nn.avg_pool, tf.nn.batch_normalization, tf.nn.bias_add, tf.nn.conv1d, tf.nn.conv2d, tf.nn.conv2d_transposed, tf.nn.conv3d, tf.nn.depthwise_conv2d, tf.nn.elu, tf.nn.embedding_lookup, tf.nn.fused_batch_norm, tf.nn.gelu, tf.nn.l2_normalize, tf.nn.leaky_relu, tf.nn.local_response_normalization, tf.nn.max_pool, tf.nn.max_pool_with_argmax, tf.nn.max_pool3d, tf.nn.moments, tf.nn.relu, tf.nn.relu6, tf.nn.rnn_cell_GRUCell, tf.nn.rnn_cell_LSTMCell, tf.nn.dynamic_rnn, tf.nn.sigmoid, tf.nn.softmax, tf.nn.swish, tf.nn.tanh, tf.not_equal, tf.one_hot, tf.pad, tf.pow, tf.realdiv, tf.reduce_any, tf.reduce_max, tf.reduce_mean, tf.reduce_sum, tf.repeat, tf.reshape, tf.expand_dims, tf.squeeze, tf.reverse, tf.reverse_sequence, tf.round, tf.rsqrt, tf.scatter_nd, tf.select, tf.signal.frame, tf.slice, tf.space_to_batch_nd, tf.space_to_depth, tf.split, tf.sqrt, tf.square, tf.stack, tf.strided_slice, tf.subtract, tf.tile, tf.transpose, tf.unstack, tf.where, tf.image.crop_and_resize, tf.image.resize_bilinear, tf.image.resize_nearest_neighbor
TensorFlow Lite
TF 2.15.0
ABS, ADD, ADD_N, ARG_MAX, ARG_MIN, AVERAGE_POOL_2D, BATCH_MATMUL, BATCH_TO_SPACE_ND, BROADCAST_ARGS, BROADCASTTO, CONCATENATION, CONV_2D, CONV_3D, CUMSUM, DEPTHWISE_CONV_2D, DEPTH_TO_SPACE, DEQUANTIZE, DIV, ELU, EMBEDDING_LOOKUP, EQUAL, EXP, EXPAND_DIMS, FLOOR, FLOOR_DIV, FLOOR_MOD, FULLY_CONNECTED, GATHER, GATHER_ND, GREATER, GREATER_EQUAL, HARD_SWISH, L2_NORMALIZATION, L2_POOL_2D, LEAKY_RELU, LESS, LESS_EQUAL, LOCAL_RESPONSE_NORMALIZATION, LOG_SOFTMAX, LOGICAL_AND, LOGICAL_NOT, LOGICAL_OR, LOGISTIC, LSTM, MAX_POOL_2D, MAXIMUM, MEAN, MIRROR_PAD, MUL, NEG, NON_MAX_SUPPRESSION_V5, NOT_EQUAL, ONE_HOT, PACK, PAD, PADV2, POW, PRELU, RANGE, RANK, REDUCE_ANY, REDUCE_MAX, REDUCE_MIN, REDUCE_PROD, RELU, RELU1, RELU6, RELU_N1_TO_1, RESIZE_BILINEAR, RESIZE_NEAREST_NEIGHBOR, REVERSE_SEQUENCE, REVERSE_V2, ROUND, RSQRT, SCATTER_ND, SEGMENT_SUM, SELECT, SHAPE, SIN, SLICE, SOFTMAX, SPACE_TO_BATCH_ND, SPACE_TO_DEPTH, SPARSE_TO_DENSE, SPLIT, SPLIT_V, SQUARE, SQUARED_DIFFERENCE, SQUEEZE, RESHAPE, STRIDED_SLICE, SUB, SUM, SVDF, TANH, TILE, TOPK, TOPK_V2, TRANSPOSE, TRANSPOSE_CONV, UNIDIRECTIONAL_SEQUENCE_LSTM, UNIQUE, UNPACK, UNSTACK, WHERE, ZEROS_LIKE
Keras
TF 2.15.0
Activation(leaky_relu), Activation(relu), Activation(sigmoid), Activation(softmax), Activation(tanh), ActivityRegularization, Add, AdditiveAttention, AlphaDropout, Attention, Average, AveragePooling1D, AveragePooling3D, BatchNormalization, BatchNormalizationV1, Concatenate, Conv1D, Conv1DTranspose, Conv2D, Conv2DTranspose, Conv3D, ConvLSTM2D, Cropping1D, Cropping2D, Cropping3D, Dense, DepthwiseConv2D, ELU, Embedding, Flatten, GRU, LayerNormalization, LeakyRelu, LSTM, Maximum, MaxPooling1D, MaxPooling2D, AveragePooling2D, GlobalAveragePooling2D, GlobalMaxPooling2D, MaxPooling3D, Minimum, Multiply, Permute, PRelu, RELU, Reshape, RNN, SeparableConv1D, SeparableConv2D, SimpleRNN, Softmax, Subtract, ThresholdedReLU, Upsampling1D, UpSampling2D, UpSampling3D, ZeroPadding1D, ZeroPadding2D, ZeroPadding3D
Caffe
absval, axpy, batchnorm, bn, concat, convolution, convolutiondepthwise, deconvolution, depthwiseconvolution, dropout, eltwise, elu, flatten, innerproduct, l2normalizescale, leakyrelu, lrn, lstm, normalize, padchannel, permute, pooling, poolwithargmax, prelu, priorbox, proposal, relu, reorg, reshape, reverse, roipooling, scale, shufflechannel, sigmoid, slice, softmax, swish, tanh
Darknet
avgpool, batchnormalize, concat, convolutional, crop, dropout, eltwise, flatten, leakyrelu, maxpool, mish, relu, route, shortcut, sigmoid, softmax, tanh
PyTorch
2.2.2（导出为 ONNX 后使用，无独立算子映射）