Neural Network Structures Supported by Renesas Translator

The translator (free version) supports microcontrollers with comparatively small ROM/RAM capacity. In order to compress the capacity used by the library, only functions that are often used by neural networks are supported.


Translatable Functions (Inference)

Neural net functions that are used only for learning and not for inference (for example, tf.nn.l2_loss for evaluating loss) are usable.

Neural Network Function TensorFlow Caffe
Convolution tf.nn.conv2d Convolution
Deconvolution tf.nn.conv2d_transpose Deconvolution
Max Pooling tf.nn.max_pool Pooling
tf.contrib.layers.max_pool2d with pooling_param {pool: MAX}
Average Pooling tf.nn.avg_pool Pooling
tf.contrib.layers.avg_pool2d with pooling_param {pool: AVE}
Relu tf.nn.relu ReLU
Tanh tf.tanh TanH
Sigmoid tf.sigmoid Sigmoid
Softmax tf.nn.softmax Softmax
Innerproduct tf.matmul InnerProduct
Other Functions tf.add Input
tf.nn.bias_add Reshape
tf.contrib.layers.bias_add Split

Supported Neural Network Structures

Convolutional Neural Network
LeNet Supported  
AlexNet Partially supported LRN functions are not supported. CNN without LRN is supported.
VGG Supported Due to the large number of FC nodes, a very large memory is required for performing conversion.
Network in Network GoogleNet Not supported DepthConcat is not supported.
Recurrent Neural Network
  Not supported Networks with built-in memory are not supported. Consider CNN as an alternative.
Long Short Term Memory
  Not supported Networks with built-in memory are not supported. Consider CNN as an alternative.
Auto Encorder   Supported Transpose functions are not supported.