1 Details about Extended Touchdown Dataset 1.1 Extended Touchdown

We build a new extended dataset of the Touchdown, which are collected by the same way as the Touchdown. First, we choose some panorama IDs in the test data of the Touchdown dataset and download the panoramasin equirectangular projection. Then we slice each into eight images and project them to perspective projection. In addition, these data are collected from the New Y ork StreetView. Figure 1: The word frequency and the length of language descriptions on the Touchdown as well as the extended Touchdown. This part shows the successful examples of SIRI and LingUnet.