Apply successivly BatchNormalization, ReLu nonlinearity, Convolution and Dropout (if dropout_p > 0) on the inputs

Loop on the batches

Load the weights

Print a summary of the network architecture

Returns the intersection I and union U (to compute the jaccard I/U) and the accuracy. :param y_pred: tensor of predictions. shape (b*0*1, c

Performs 1x1 convolution followed by softmax nonlinearity The output will have the shape (batch_size * n_rows * n_cols, n_classes)

Apply first a BN_ReLu_conv layer with filter size = 1, and a max pooling with a factor 2

Performs upsampling on block_to_upsample by a factor 2 and concatenates it with the skip_connection

Save the weights

This function builds the model defined in config_path and restores the weights defined in weight_path. It then reports the jaccard and global

This code implements the Fully Convolutional DenseNet described in https://arxiv.org/abs/1611.09326 The network consist of a downsamp

We only display the layers in the list below

Similar to theano_metrics to metrics but instead y_pred and y_true are now numpy arrays