Open unizard opened 6 years ago
Figure 1: Diverse answers generated by FLIPDIAL in the one-way visual dialogue (1VD) task. For a given time step (row), each column shows a generated answer to the current question. Answers are obtained by decoding a latent zi sampled from the conditional prior – with conditions being image, caption and dialogue history up until that time step.
Institute: Oxford, Torr URL: https://arxiv.org/pdf/1802.03803.pdf Web: Category: Visual Dialog Interest: 3