How to align scene_ids in data_part2_scene.json from those in ScanNet?

[ch3cook-fdu]

I notice that the scene_ids in data_part2_scene.json have a dtype of int, while the scene_ids in scannet are something like scene0000_00. So, how do I align these two datasets? Or how can I split the scannet annotations from that file by myself?

[evelinehong]

please refer to https://drive.google.com/drive/folders/1bJpK00UFRZGH7tdx3PaqDrrTfHNP6B2K?usp=drive_link

[cy94]

Hello,
I used the final_scene_map_dict_scan_v2.json file but it doesnt contain scene ID mappings for all the annotations in data_part2_scene_v2.json. Is there are a different scene ID mapping that I can use?

[ch3cook-fdu]

Hello, I used the final_scene_map_dict_scan_v2.json file but it doesnt contain scene ID mappings for all the annotations in data_part2_scene_v2.json. Is there are a different scene ID mapping that I can use?

I think the scannet part is only part of data_part2_scene_v2.json. However, the annotations is quite noisy, and contain many duplicates...

[cy94]

Thanks! Out of 202k train annotations about 60k could be matched with ScanNet IDs, is that correct? and are the rest from Habitat-3D?

[ch3cook-fdu]

Thanks! Out of 202k train annotations about 60k could be matched with ScanNet IDs, is that correct? and are the rest from Habitat-3D?

I think the precise count would be 57,434?

[cy94]

I found 60582 in data_part2_scene_v2.json, is that the right file to use?

[ch3cook-fdu]

I am using the following code for filtering:

import json
mapping_file = json.load(open('final_scene_map_dict_scan_v2.json', 'r'))
data = json.load(open('data_part2_scene.json', 'r'))
scannet_annotations = []
corpus = set()
for anno in data:
    scannet_scene_id = mapping_file.get(str(anno['scene_id']), None)
    if scannet_scene_id is None:
        continue
    scannet_annotations.append(anno)
print(len(scannet_annotations))

[cy94]

I see, I'm using data_part2_scene_v2.json, I didn't see v1 (data_part2_scene.json) anywhere on your Google Drive. Maybe thats why there are more?

[ch3cook-fdu]

That might be the case!