tensorflow implement of bounding box to mask transform

tf_bbox2mask.py

## ref: https://stackoverflow.com/questions/34128104/tensorflow-creating-mask-of-varied-lengths
def tf_bbox2mask(y1, x1, y2, x2, img_H, img_W):
    ## Repeat for each row or column
    y1_transposed = tf.expand_dims(tf.tile(y1,[img_W]), 0)
    x1_transposed = tf.expand_dims(tf.tile(x1,[img_H]), 1)
    y2_transposed = tf.expand_dims(tf.tile(y2,[img_W]), 0)
    x2_transposed = tf.expand_dims(tf.tile(x2,[img_H]), 1)
    ## Get the range grid
    range_row = tf.cast(tf.expand_dims(tf.range(0, img_H, 1), 1), tf.int32)
    range_col = tf.cast(tf.expand_dims(tf.range(0, img_W, 1), 0), tf.int32)
    ## Generate bollean masks
    mask_y1 = tf.less(y1_transposed, range_row)
    mask_x1 = tf.less(x1_transposed, range_col)
    mask_y2 = tf.less(range_row, y2_transposed)
    mask_x2 = tf.less(range_col, x2_transposed)

    result = tf.to_float(mask_y1)*tf.to_float(mask_x1)*tf.to_float(mask_y2)*tf.to_float(mask_x2)
    return result