18 KiB
18 KiB
INFO: Actual implementations may differ from reference link in terms of dimensions and parameters supported
| TIM-VX API | Internal Op | Status | Reference |
|---|---|---|---|
| Add | ADD | Mapped | tf.math.add |
| Multiply | MULTIPLY | Mapped | tf.math.multiply |
| Conv2d | CONV2D | Mapped | tf.nn.conv2d tf.nn.atros_conv2d tf.nn.depthwise_conv2d |
| Softmax | SOFTMAX | Mapped | tf.nn.softmax |
| Pool2d | POOL | Mapped | tf.nn.pool |
| LeakyRelu | LEAKY_RELU | Mapped | tf.nn.leaky_relu |
| Concat | CONCAT | Mapped | tf.concat |
| Split | SPLIT | Mapped | tf.split |
| BatchNorm | BATCH_NORM | Mapped | tf.nn.batch_normalization |
| DeConv2d | DECONVOLUTION | Mapped | tf.nn.conv2d_transpose |
| Reshape | RESHAPE | Mapped | tf.reshape |
| Transpose | PERMUTE | Mapped | tf.transpose |
| Prelu | PRELU | Mapped | tf.keras.layers.PReLU |
| MaxUnpool2d | UPSAMPLE | Mapped | tfa.layers.MaxUnpooling2D |
| Relu | RELU | Mapped | tf.nn.relu |
| Reorg | REORG | Mapped | darknet.reorg |
| L2Normalization | L2_NORMALIZE | Mapped | tf.math.l2_normalize |
| FullyConnected | FCL2 | Mapped | tf.keras.layers.Dense |
| MaxpoolWithArgmax | POOLWITHARGMAX | Mapped | tf.nn.max_pool_with_argmax |
| ArgMax | ARGMAX | Mapped | tf.math.argmax |
| Maximum | MAXIMUM | Mapped | tf.math.maximum |
| Sub | SUBTRACT | Mapped | tf.math.subtract |
| Relu6 | RELU6 | Mapped | tf.nn.relu6 |
| Sigmoid | SIGMOID | Mapped | tf.math.sigmoid |
| Tanh | TANH | Mapped | tf.math.tanh |
| Sqrt | SQRT | Mapped | tf.math.sqrt |
| Rsqrt | RSQRT | Mapped | tf.math.rsqrt |
| SoftRelu | SOFTRELU | Mapped | tf.math.softplus |
| Div | DIVIDE | Mapped | tf.math.divide |
| Dropout | DROPOUT | Mapped | f(x) = x*ratio |
| Resize | RESIZE | Mapped | tf.image.resize |
| Reverse | REVERSE | Mapped | tf.reverse |
| DepthToSpace | DEPTH2SPACE | Mapped | tf.nn.depth_to_space |
| SpaceToDepth | SPACE2DEPTH | Mapped | tf.nn.space_to_depth |
| DataConvert | DATACONVERT | Mapped | Data Format Conversion |
| Slice | SLICE | Mapped | tf.slice |
| Elu | ELU | Mapped | tf.nn.elu |
| Batch2Space | BATCH2SPACE | Mapped | tf.batch_to_space |
| Space2Batch | SPACE2BATCH | Mapped | tf.space_to_batch |
| Pad | PAD | Mapped | tf.pad |
| Matmul | MATRIXMUL | Mapped | tf.linalg.matmul |
| LayerNormalization | LAYER_NORM | Mapped | tf.keras.layers.LayerNormalization |
| ReduceMin | REDUCE_MIN | Mapped | tf.math.reduce_min |
| ReduceMax | REDUCE_MAX | Mapped | tf.math.reduce_max |
| ReduceAny | REDUCE_ANY | Mapped | tf.math.reduce_any |
| ReduceProd | REDUCE_PROD | Mapped | tf.math.reduce_prod |
| ReduceMean | REDUCE_MEAN | Mapped | tf.math.reduce_mean |
| InstanceNormalization | INSTANCE_NORM | Mapped | tfa.layers.InstanceNormalization |
| StridedSlice | STRIDED_SLICE | Mapped | tf.strided_slice |
| Abs | ABS | Mapped | tf.math.abs |
| Conv1d | CONV1D | Mapped | tf.nn.conv1d |
| NBG | NBG | Mapped | Network Binary Graph |
| LocalResponseNormalization | LRN2 | Mapped | tf.nn.local_response_normalization |
| Greater | RELATIONAL_OPS_GREATER | Mapped | tf.math.greater |
| GreaterOrEqual | RELATIONAL_OPS_GREATER_EQUAL | Mapped | tf.math.greater_equal |
| Less | RELATIONAL_OPS_LESS | Mapped | tf.math.less |
| LessOrEqual | RELATIONAL_OPS_LESS_EQUAL | Mapped | tf.math.less_equal |
| Equal | RELATIONAL_OPS_EQUAL | Mapped | tf.math.equal |
| NotEqual | RELATIONAL_OPS_NOT_EQUAL | Mapped | tf.math.not_equal |
| Pow | POW | Mapped | tf.math.pow |
| FloorDiv | FLOORDIV | Mapped | tf.math.floordiv |
| Minimum | MINIMUM | Mapped | tf.math.minimum |
| And | LOGICAL_OPS | Mapped | tf.math.logical_and |
| Or | LOGICAL_OPS | Mapped | tf.math.logical_or |
| Select | SELECT | Mapped | tf.where |
| Relu1 | RELU1 | Mapped | tf.keras.layers.ReLU(max_value=1.0) |
| Stack | STACK | Mapped | tf.stack |
| Floor | FLOOR | Mapped | tf.math.floor |
| Square | SQUARE | Mapped | tf.math.square |
| Neg | NEG | Mapped | tf.math.negative |
| Exp | EXP | Mapped | tf.math.exp |
| Clip | CLIP | Mapped | tf.clip_by_value |
| AddN | ADDN | Mapped | tf.math.add_n |
| Gather | GATHER | Mapped | tf.gather |
| LogicalNot | LOGICAL_NOT | Mapped | tf.math.logical_not |
| Sin | SIN | Mapped | tf.math.sin |
| Log | LOG | Mapped | tf.math.log |
| ArgMin | ARGMIN | Mapped | tf.math.argmin |
| LogSoftmax | LOG_SOFTMAX | Mapped | tf.nn.log_softmax |
| Swish | SWISH | Mapped | tf.keras.activations.swish |
| HardSwish | SWISH | Mapped | torch.nn.Hardswish |
| GatherNd | GATHER_ND | Mapped | tf.gather_nd |
| Cast | CAST | Mapped | tf.cast |
| Moments | MOMENTS | Mapped | tf.moments |
| Squeeze | SQUEEZE | Mapped | tf.squeeze |
| HardSigmoid | HARD_SIGMOID | Mapped | tf.keras.activations.hard_sigmoid |
| Mish | MISH | Mapped | tfa.activations.mish |
| DeConv1d | DECONVOLUTION1D | Mapped | tf.nn.conv1d_transpose |
| Resize1d | RESIZE_1D | Mapped | Onnx.resize 1D image |
| Linear | LINEAR | Mapped | tf.keras.activations.linear |
| ScatterND | SCATTER_ND | Mapped | tf.scatter_nd |
| Unstack | UNSTACK | Mapped | tf.unstack |
| Tile | TILE | Mapped | tf.tile |
| GroupedConv2d | GROUPED_CONV2D | Mapped | ANEURALNETWORKS_GROUPED_CONV_2D |
| SpatialTransformer | SPATIAL_TRANSFORMER | Mapped | SpatialTransformer |
| PROPOSAL | Planned 21Q3 | Faster-RCNN Proposal Layer | |
| ROI_POOL | Planned 21Q3 | ANEURALNETWORKS_ROI_POOLING | |
| ROI_ALIGN | Planned 21Q3 | ANEURALNETWORKS_ROI_ALIGN | |
| SHUFFLECHANNEL | Planned 21Q3 | ANEURALNETWORKS_CHANNEL_SHUFFLE | |
| SIGNAL_FRAME | Planned 21Q3 | tf.signal.frame | |
| TOPK | Planned 21Q3 | tf.math.top_k | |
| GRUCell | GRUCELL_OVXLIB | Planned 21Q3 | tf.keras.layers.GRUCell |
| UnidirectionalSequenceGRU | GRU_OVXLIB | Planned 21Q4 | tf.keras.layers.GRU |
| UnidirectionalSequenceRNN | UNIDIRECTIONAL_SEQUENCE_RNN | Planned 21Q4 | ANEURALNETWORKS_UNIDIRECTIONAL_SEQUENCE_RNN |
| BidirectionalSequenceRNN | BIDIRECTIONAL_SEQUENCE_RNN | Planned 21Q4 | ANEURALNETWORKS_BIDIRECTIONAL_SEQUENCE_RNN |
| RNNCell | RNNCELL_OVXLIB | Planned 21Q3 | ANEURALNETWORKS_RNN |
| BidirectionalSequenceLSTM | BIDIRECTIONAL_SEQUENCE_LSTM | Planned 21Q4 | ANEURALNETWORKS_BIDIRECTIONAL_SEQUENCE_LSTM |
| UnidirectionalSequenceLSTM | LSTM_OVXLIB | Mapped | ANEURALNETWORKS_UNIDIRECTIONAL_SEQUENCE_LSTM |
| LSTMCell | LSTMUNIT_OVXLIB | replace with UnidirectionalSequenceLSTM by set n_step = 1 | ANEURALNETWORKS_LSTM |
| PRE_PROCESS | Planned 21Q4 | Image Preprocessing (YUV2RGB, Input Normalization, Resizing, etc) | |
| HASHTABLE_LOOKUP | Planned 21Q4 | ANEURALNETWORKS_HASHTABLE_LOOKUP | |
| EMBEDDING_LOOKUP | Planned 21Q4 | ANEURALNETWORKS_EMBEDDING_LOOKUP | |
| LSH_PROJECTION | Planned 21Q4 | ANEURALNETWORKS_LSH_PROJECTION | |
| SVDF | Planned 21Q4 | ANEURALNETWORKS_SVDF | |
| HEATMAP_MAX_KEYPOINT | Planned 21Q4 | ANEURALNETWORKS_HEATMAP_MAX_KEYPOINT | |
| AXIS_ALIGNED_BBOX_TRANSFORM | Planned 21Q4 | ANEURALNETWORKS_AXIS_ALIGNED_BBOX_TRANSFORM | |
| BOX_WITH_NMS_LIMIT | Planned 21Q4 | ANEURALNETWORKS_BOX_WITH_NMX_LIMIT | |
| GENERATE_PROPOSALS | Planned 21Q4 | ANEURALNETWORKS_GENERATE_PROPOSALS | |
| DETECTION_POSTPROCESS | Planned 21Q4 | ANEURALNETWORKS_DETECTION_POSTPROCESSING | |
| RANDOM_MULTINOMIAL | Planned 21Q4 | ANEURALNETWORKS_RANDOM_MULTINOMIAL | |
| UPSAMPLESCALE | Planned 21Q4 | Unoffical Caffe layer | |
| GROUP_NORM | Planned 21Q4 | tfa.layers.GroupNormalization | |
| ROUND | Planned 21Q4 | tf.math.round | |
| CEIL | Planned 21Q4 | tf.math.ceil | |
| SEQUENCE_MASK | Planned 21Q4 | tf.math.ceil | |
| REPEAT | Planned 21Q4 | tf.repeat | |
| ERF | Planned 21Q4 | tf.math.erf | |
| ONE_HOT | Planned 21Q4 | tf.one_hot | |
| NMS | Planned 21Q4 | tf.image.non_max_suppression | |
| GROUPED_CONV1D | Planned 21Q4 | ||
| SCATTER_ND_UPDATE | Planned 21Q4 | tf.compat.v1.scatter_nd_update | |
| GELU | Planned 21Q4 | tf.nn.gelu | |
| CONV_RELU | Deprecated | ||
| CONV_RELU_POOL | Deprecated | ||
| FCL | Deprecated | ||
| FCL_RELU | Deprecated | ||
| LRN | Deprecated | ||
| SCALE | Deprecated | ||
| DEPTHWISE_CONV1D | Deprecated | ||
| L2NORMALIZESCALE | Deprecated | ||
| INTERP | Deprecated | ||
| NOOP | Deprecated | ||
| TENSORSTACKCONCAT | Deprecated | ||
| VARIABLE | InternalOnly | ||
| RELUN | Deprecated | ||
| CROP | Deprecated | ||
| TENSOR_ADD_MEAN_STDDEV_NORM | InternalOnly | ||
| RNN | Deprecated | ||
| LSTMUNIT_ACTIVATION | InternalOnly | ||
| LSTM | Deprecated | ||
| LSTMUNIT | Deprecated | ||
| QUANTIZED_16BIT_LSTM | InternalOnly | ||
| RELU_KERAS | Deprecated | ||
| PRE_PROCESS_GRAY | InternalOnly | ||
| PRE_PROCESS_YUV444 | InternalOnly | ||
| PRE_PROCESS_NV12 | InternalOnly | ||
| PRE_PROCESS_YUV420 | InternalOnly | ||
| PRE_PROCESS_BGRA | InternalOnly | ||
| PRE_PROCESS_TENSOR | InternalOnly | ||
| IMAGEPROCESS | Deprecated | ||
| POST_PROCESS | InternalOnly | ||
| EXTRA_ENDING | InternalOnly | ||
| SYNC_HOST | InternalOnly | ||
| BATCHNORM_SINGLE | InternalOnly | ||
| CONCATSHIFT | InternalOnly | ||
| A_TIMES_B_PLUS_C | Deprecated | tf.add(tf.mul(A, B), C) |