About the label ID

[Hiusam]

Hi, thanks for your great work!

Could you please give me some hints about where self.type2class and self.nyu40ids come from? I check scannetv2-labels.combined.tsv and they seem not well aligned.

class ScannetDatasetConfig:

    def __init__(self, num_class=485, agnostic=False):
        self.num_class = num_class if not agnostic else 1  # 18
        self.num_heading_bin = 1
        self.num_size_cluster = num_class
        if num_class == 18:
            self.type2class = {'cabinet': 0, 'bed': 1, 'chair': 2, 'couch': 3, 'table': 4, 'door': 5,
                               'window': 6, 'bookshelf': 7, 'picture': 8, 'counter': 9, 'desk': 10, 'curtain': 11,
                               'refrigerator': 12, 'shower curtain': 13, 'toilet': 14, 'sink': 15, 'bathtub': 16,
                               'other furniture': 17}
        else:
            self.type2class = {'wall': 0, 'chair': 1, 'floor': 2, ...}

        self.class2type = {self.type2class[t]: t for t in self.type2class}
        if num_class == 18:
            self.nyu40ids = np.array([3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 16, 24, 28, 33, 34, 36, 39])
        else:
            # only train
            self.nyu40ids = np.array([1, 2, 3, 4, 5, 6, 7,... ])
        self.nyu40id2class = {nyu40id: i for i, nyu40id in enumerate(list(self.nyu40ids))}

Also, how did you get the cls_results.json?

[nickgkan]

The nyu ids come from the file you mentioned, but if you notice in joint_dataset.py we handle the 18 and the 485 cases differently, i.e., use different id column.

The cls_results.json contains predicted classes from a pointnet++ we trained.