TensorFlow utils

check_checkpoints.py

from tensorflow.python import pywrap_tensorflow

# Get weights or whatever you want.
# Notice the filename should be full pathname, e.g. './model-7000'
def get_weights(filename):
    reader = pywrap_tensorflow.NewCheckpointReader(filename)
    var_to_shape_map = reader.get_variable_to_shape_map()
    weights = [reader.get_tensor(key) for key in var_to_shape_map if 'weights' in key]
    return weights

dice_loss.py

def dice_loss(y_true, y_conv):
    """Compute dice among **positive** labels to avoid unbalance.

    Argument:
        y_true: [batch_size, depth, height, width, 1]
        y_conv: [batch_size, depth, height, width, 2]
    """
    y_true = tf.to_float(tf.reshape(y_true[..., 0], [-1]))
    y_conv = tf.to_float(tf.reshape(y_conv[..., 1], [-1]))
    intersection = tf.reduce_sum(y_conv * y_true)
    union = tf.reduce_sum(y_conv * y_conv) + tf.reduce_sum(y_true * y_true)  # y_true is binary
    dice_coef = 2.0 * intersection / union
    return 1 - tf.clip_by_value(dice_coef, 0, 1.0 - 1e-7)