Closed ir0 closed 7 years ago
Hi ir0,
This is a good point, unfortunately I do not have the mapping to the extended 37 classes, but the way I produced that dictionary can now be seen in the repo. There is now a folder called semantic_classes
with a script semantic_class_dictionary_generator.py
. It produces the dictionary from a textfile in the same folder called wnid_to_class.txt
which I have also added to the repo.
In wnid_to_class.txt
we manually mapped each of the wnids (using their plain english name) to the NYU 13 classes. If you add a new column to that textfile with a custom mapping, and update the column_name
and class_list
variables in the semantic_class_dictionary_generator.py
script, as well as giving the actual list of classes in order of their id (i.e. a NYU_37_CLASSES
) it will produce the corresponding dictionary.
Hope this is helpful, sorry we don't have the exact mapping to hand.
Thanks, that's helpful.
Hello @ir0 , Did you end up implementing the 37 classes? Also, how did you get the number 37 from?
I just downloaded the NYUDv2 dataset from link and inside the mat, it looks like there are 894 classes?
Look for those here https://github.com/ankurhanda/SceneNetv1.0/ under the section "Conversion from SUN RGB-D/NYUv2 37/40 Labels"
Hi John,
I see you have the mapping of the WordNet IDs to the 13 NYU classes. Do you have the mapping to the extended 37 classes or, if not, how I could easily get it?
Thanks!