src/armnnTfParser/TensorFlowSupport.md

   1 # TensorFlow operators that the Arm NN SDK supports
   2
   3 This reference guide provides a list of TensorFlow operators the Arm NN SDK currently supports.
   4
   5 The Arm NN SDK TensorFlow parser currently only supports fp32 operators.
   6
   7 ## Fully supported
   8
   9 **avg_pool**
  10
  11 See the TensorFlow [avg_pool documentation](https://www.tensorflow.org/api_docs/python/tf/nn/avg_pool) for more information.
  12
  13 **bias_add**
  14
  15 See the TensorFlow [bias_add documentation](https://www.tensorflow.org/api_docs/python/tf/nn/bias_add) for more information.
  16
  17 **conv2d**
  18
  19 See the TensorFlow [conv2d documentation](https://www.tensorflow.org/api_docs/python/tf/nn/conv2d) for more information.
  20
  21 **expand_dims**
  22
  23 See the TensorFlow [expand_dims documentation](https://www.tensorflow.org/api_docs/python/tf/expand_dims) for more information.
  24
  25 **gather**
  26
  27 See the TensorFlow [gather documentation](https://www.tensorflow.org/api_docs/python/tf/gather) for more information.
  28
  29 **identity**
  30
  31 See the TensorFlow [identity documentation](https://www.tensorflow.org/api_docs/python/tf/identity) for more information.
  32
  33 **local_response_normalization**
  34
  35 See the TensorFlow [local_response_normalization documentation](https://www.tensorflow.org/api_docs/python/tf/nn/local_response_normalization)  for more information.
  36
  37 **max_pool**
  38
  39 See the TensorFlow [max_pool documentation](https://www.tensorflow.org/api_docs/python/tf/nn/max_pool) for more information.
  40
  41 **placeholder**
  42
  43 See the TensorFlow [placeholder documentation](https://www.tensorflow.org/api_docs/python/tf/placeholder) for more information.
  44
  45 **reduce_mean**
  46
  47 See the TensorFlow [reduce_mean documentation](https://www.tensorflow.org/api_docs/python/tf/reduce_mean) for more information.
  48
  49 **relu**
  50
  51 See the TensorFlow [relu documentation](https://www.tensorflow.org/api_docs/python/tf/nn/relu) for more information.
  52
  53 **relu6**
  54
  55 See the TensorFlow [relu6 documentation](https://www.tensorflow.org/api_docs/python/tf/nn/relu6) for more information.
  56
  57 **rsqrt**
  58
  59 See the TensorFlow [rsqrt documentation](https://www.tensorflow.org/api_docs/python/tf/math/rsqrt) for more information.
  60
  61 **shape**
  62
  63 See the TensorFlow [shape documentation](https://www.tensorflow.org/api_docs/python/tf/shape) for more information.
  64
  65 **sigmoid**
  66
  67 See the TensorFlow [sigmoid documentation](https://www.tensorflow.org/api_docs/python/tf/sigmoid) for more information.
  68
  69 **softplus**
  70
  71 See the TensorFlow [softplus documentation](https://www.tensorflow.org/api_docs/python/tf/nn/softplus) for more information.
  72
  73 **squeeze**
  74
  75 See the TensorFlow [squeeze documentation](https://www.tensorflow.org/api_docs/python/tf/squeeze) for more information.
  76
  77 **tanh**
  78
  79 See the TensorFlow [tanh documentation](https://www.tensorflow.org/api_docs/python/tf/tanh) for more information.
  80
  81 **transpose**
  82
  83 See the TensorFlow [transpose documentation](https://www.tensorflow.org/api_docs/python/tf/transpose) for more information.
  84
  85 ## Partially supported
  86
  87 **add**
  88
  89 The parser does not support all forms of [broadcast composition](https://www.tensorflow.org/performance/xla/broadcasting), only broadcasting of scalars and 1D tensors. See the TensorFlow [add operator documentation](https://www.tensorflow.org/api_docs/python/tf/add) for more information.
  90
  91 **add_n**
  92
  93 The parser does not support all forms of [broadcast composition](https://www.tensorflow.org/performance/xla/broadcasting), only broadcasting of scalars and 1D tensors. See the TensorFlow [add operator documentation](https://www.tensorflow.org/api_docs/python/tf/add_n) for more information.
  94
  95 **concat**
  96
  97 Arm NN supports concatenation along the channel dimension for data formats NHWC and NCHW.
  98
  99 **constant**
 100
 101 The parser does not support the optional `shape` argument. It always infers the shape of the output tensor from `value`. See the TensorFlow [constant documentation](https://www.tensorflow.org/api_docs/python/tf/constant) for further information.
 102
 103 **depthwise_conv2d_native**
 104
 105 The parser only supports a dilation rate of (1,1,1,1). See the TensorFlow [depthwise_conv2d_native documentation](https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d_native) for more information.
 106
 107 **equal**
 108
 109 The parser does not support all forms of [broadcast composition](https://www.tensorflow.org/performance/xla/broadcasting), only broadcasting of 4D and 1D tensors. See the TensorFlow [equal operator documentation](https://www.tensorflow.org/api_docs/python/tf/math/equal) for more information.
 110
 111 **fused_batch_norm**
 112
 113 The parser does not support training outputs. See the TensorFlow [fused_batch_norm documentation](https://www.tensorflow.org/api_docs/python/tf/nn/fused_batch_norm) for more information.
 114
 115 **greater**
 116
 117 The parser does not support all forms of [broadcast composition](https://www.tensorflow.org/performance/xla/broadcasting), only broadcasting of 4D and 1D tensors. See the TensorFlow [greater operator documentation](https://www.tensorflow.org/api_docs/python/tf/math/greater) for more information.
 118
 119 **matmul**
 120
 121 The parser only supports constant weights in a fully connected layer. See the TensorFlow [matmul documentation](https://www.tensorflow.org/api_docs/python/tf/matmul) for more information.
 122
 123 **maximum**
 124
 125 where maximum is used in one of the following ways
 126
 127 * max(mul(a, x), x)
 128 * max(mul(x, a), x)
 129 * max(x, mul(a, x))
 130 * max(x, mul(x, a)
 131
 132 This is interpreted as a ActivationLayer with a LeakyRelu activation function. Any other usage of max will result in the insertion of a simple maximum layer. The parser does not support all forms of [broadcast composition](https://www.tensorflow.org/performance/xla/broadcasting). See the TensorFlow [maximum documentation](https://www.tensorflow.org/api_docs/python/tf/maximum) for more information.
 133
 134 **minimum**
 135
 136 The parser does not support all forms of [broadcast composition](https://www.tensorflow.org/performance/xla/broadcasting), only broadcasting of 4D and 1D tensors. See the TensorFlow [minimum operator documentation](https://www.tensorflow.org/api_docs/python/tf/math/minimum) for more information.
 137
 138 **multiply**
 139
 140 The parser does not support all forms of [broadcast composition](https://www.tensorflow.org/performance/xla/broadcasting), only broadcasting of scalars and 1D tensors. See the TensorFlow [multiply documentation](https://www.tensorflow.org/api_docs/python/tf/multiply) for more information.
 141
 142 **pad**
 143
 144 Only supports tf.pad function with mode = 'CONSTANT' and constant_values = 0. See the TensorFlow [pad documentation](https://www.tensorflow.org/api_docs/python/tf/pad) for more information.
 145
 146 **realdiv**
 147
 148 The parser does not support all forms of [broadcast composition](https://www.tensorflow.org/performance/xla/broadcasting), only broadcasting of scalars and 1D tensors. See the TensorFlow [realdiv documentation](https://www.tensorflow.org/api_docs/python/tf/realdiv) for more information.
 149
 150 **reshape**
 151
 152 The parser does not support reshaping to or from 4D. See the TensorFlow [reshape documentation](https://www.tensorflow.org/api_docs/python/tf/reshape) for more information.
 153
 154 **resize_images**
 155
 156 The parser only supports `ResizeMethod.BILINEAR` with `align_corners=False`. See the TensorFlow [resize_images documentation](https://www.tensorflow.org/api_docs/python/tf/image/resize_images) for more information.
 157
 158 **softmax**
 159
 160 The parser only supports 2D inputs and does not support selecting the `softmax` dimension. See the TensorFlow [softmax documentation](https://www.tensorflow.org/api_docs/python/tf/nn/softmax) for more information.
 161
 162 **split**
 163
 164 Arm NN supports split along the channel dimension for data formats NHWC and NCHW.
 165
 166 **subtract**
 167
 168 The parser does not support all forms of broadcasting [broadcast composition](https://www.tensorflow.org/performance/xla/broadcasting), only broadcasting of scalars and 1D tensors. See the TensorFlow [subtract documentation](https://www.tensorflow.org/api_docs/python/tf/math/subtract) for more information.
 169
 170 **pack/stack**
 171
 172 See the TensorFlow [stack documentation](https://www.tensorflow.org/api_docs/python/tf/stack) for more information.
 173
 174 **strided_slice**
 175
 176 See the TensorFlow [strided_slice documentation](https://www.tensorflow.org/api_docs/python/tf/strided_slice) for more information.
 177
 178 ## Tested networks
 179
 180 Arm tests these operators with the following TensorFlow fp32 neural networks:
 181
 182 * Cifar10
 183
 184 * Lenet
 185
 186 * Simple MNIST. For more information check out the [tutorial](https://developer.arm.com/technologies/machine-learning-on-arm/developer-material/how-to-guides/deploying-a-tensorflow-mnist-model-on-arm-nn) on the Arm Developer portal.
 187
 188 * mobilenet_v1_1.0_224. The Arm NN SDK only supports the non-quantized version of the network. See the [MobileNet_v1 documentation](https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet_v1.md) for more information on quantized networks.
 189
 190 * inception_v3. The Arm NN SDK only supports the official inception_v3 transformed model. See the TensorFlow documentation on [preparing models for mobile deployment](https://www.tensorflow.org/mobile/prepare_models) for more information on how to transform the inception_v3 network.
 191
 192 * ResNet v2 50 implementation from the [TF Slim model zoo](https://github.com/tensorflow/models/tree/master/research/slim)
 193
 194 More machine learning operators will be supported in future releases.