Great work. I wonder if you will release the weights trained on the stock video dataset, as described in your paper.