Inconsistency in semantic label and name

[edwardjjj]

I find some inconsistency in semantic labeling and its corresponding name.

In _preprocess_obs @ sem_exp.py, the rednet prediction use the table mp_categories_mapping to map the output labels to matterport labels. In the main loop where the LLM read these semantic labels, it uses hm3d_category to get the text back. However, these to list does not exactly match. The last entry in mp_categories_mapping corresponds to sink, but in hm3d_category, the entry is stairs.

mp_categories_mapping = [4, 11, 15, 12, 19, 23, 26, 24, 28, 38, 21, 16, 14, 6, 16]
# chair, sofa, plant, bed, toilet, tv_monitor, bathtub, shower, fireplace, appliances, towel, sink, chest_of_drawers, table, sink

hm3d_category = [
    "chair",
    "sofa",
    "plant",
    "bed",
    "toilet",
    "tv_monitor",
    "bathtub",
    "shower",
    "fireplace",
    "appliances",
    "towel",
    "sink",
    "chest_of_drawers",
    "table",
    "stairs",
]

[weizhenFrank]

Yeah, that's true. I also found such problem exists.