nickgkan / butd_detr

Code for the ECCV22 paper "Bottom Up Top Down Detection Transformers for Language Grounding in Images and Point Clouds"
Other
74 stars 11 forks source link

A potential bug of visual_data_handler.py #14

Closed Hiusam closed 1 year ago

Hiusam commented 1 year ago

Hi, in src/visual_data_handlers.py:

def load_point_cloud(self, keep_points=50000):
        """Load point-cloud information."""
        # Load labels
        label = None
        if osp.exists(self.scan_id + '_vh_clean_2.labels.ply'): # This will always be False
            data = PlyData.read(osp.join(
                self.top_scan_dir,
                self.scan_id, self.scan_id + '_vh_clean_2.labels.ply'
            ))
            label = np.asarray(data.elements[0].data['label'])

osp.exists(self.scan_id + '_vh_clean_2.labels.ply') will always be False since you seem to forget to add self.top_scan_dir and self.scan_id. This will result in label being None and will throw an error when calling self.get_object_semantic_label()

Will this affect the exp results, though no error occurred when I ran the codes?

nickgkan commented 1 year ago

Hi, at no point we use the semantic labels of the point cloud, so that's why it works. You're right, label is always None, but get_object_semantic_label is never used. We just forgot to delete that part when we cleaned the code.

Hiusam commented 1 year ago

Ok, thanks.