TensorFlow team launched the new customizable TensorBoard API on Sept 11, 2017. As the previous TensorBoard API did not include reusable APIs, it was difficult to add new visualizations to TensorBoard for users. From now on, the new TensorBoard API allows users to customize their own creation of useful visualizations and extend it to cover a wider variety of use cases.
Below is a demo video feed of data (e.g. gradients and convolution filters) as a model trains.