Reading Scene Text with Aggregated Temporal Convolutional Encoder