The number of objects
Closed this issue · 3 comments
kimkj38 commented
he-dhamo commented
Hello, this is because we work with the subset from 3DSSG (used for the learning methods introduced in the 3DSSG paper). For instance, the subset discards instances from 3RScan that are from an under-represented class, i.e. works with 160 object classes instead of around 500 in the original set. The gaps in our instance ids are there so that the correspondence with the original ids from the full 3DSSG are kept, for example object instance 37 in the graph-to-3d data is the same as object instance 37 in the full 3DSSG data. For the missing ids (e.g. 36) there is no oriented box annotation.