Visualization code

[Gorluxor]

Dear authors,

I have checked the repo for the custom visualization code but did not find anything, thus assuming something like detectron2:

from detectron2.utils.visualizer import Visualizer

meta_data = # metadata of the dataset 
data = # first image of the dataset in the detectron2 dict format

v = Visualizer(data['image'].permute(1,2,0).numpy(), metadata=meta_data, scale=1.0)
out = v.draw_sem_seg(one_data[0]['obj_part_sem_seg'])
Image.fromarray(out.get_image())

The problem is that I get a nonsense class label visualized compared to the image (using ground truth) with the code above.

Is there an example of any of the following:

1. a script that uses the inference output
2. a single inference example from image input to visualization (as seen in the paper)?