Where do you store the corresponding character names in the m3-bench-robot dataset?

[euwern]

Hi, thank you for sharing your dataset.

I try to visualize the m3-bench-robot memory graph, and I can load the memory graphs for each video, and it does have character_mappings (which stores the mapping of faces/voice to characters). Where do you store each character's name?

Do I have to manually find them in the subtitles provided in the dataset?

Please advise.

[euwern]

Furthermore, I am really confused about the character_mappings provided in the memory graph.
for instance, bedroom_01.pkl
the following are the character_mappings
{'character_0': ['voice_495', 'voice_494'], 'character_1': ['face_81', 'voice_24', 'voice_459', 'voice_25', 'voice_193', 'voice_221', 'voice_220', 'voice_219', 'voice_1194', 'face_805'], 'character_2': ['voice_58'], 'character_3': ['voice_313'], 'character_4': ['voice_348'], 'character_5': ['voice_386'], 'character_6': ['voice_737'], 'character_7': ['voice_890'], 'character_8': ['voice_891'], 'character_9': ['voice_1928']}
Why are there 10 characters in bedroom_01.mp4 video? I only spot two persons in the videos. Additionally, why are face_81 and face_805 mapped to the same character? As a human, I can see that face_81 and face_805 are clearly two different person. Am i missing any processing steps to identify individuals in the video?

Please advise.