search

EricGuo5513 / HumanML3D

HumanML3D: A large and diverse 3d human motion-language dataset.
MIT License
787 stars 78 forks source link

Where the external dictionary of language words was defined ? #30

Closed rd20karim closed 1 year ago

rd20karim commented 1 year ago

Hello, In the paper it was mentioned that an external dictionary was manually constructed to collect motion-related words and categorize them into four types: direction, body part, object and action

I can't find where it was defined in this project code ?

EricGuo5513 commented 1 year ago

Hi, it is in another project. See https://github.com/EricGuo5513/text-to-motion/blob/8640570682a0d0278b028e9525f385ee61f25ba9/utils/word_vectorizer.py#L46.

rd20karim commented 1 year ago

Thank you for providing the direct link. I see that five categories were defined, it appears that there are two different datasets - HumanML3D (22 joints) and The generated KiT version (21 joints). I'm interested in knowing if this dictionary categorizes all of the vocabulary found within these datasets.

EricGuo5513 commented 1 year ago

Yes, for both dataset we use the same dictionary.

On Wed, Jan 25, 2023 at 6:55 AM rd20karim @.***> wrote:

Thank you for providing the direct link. I see that five categories were defined, it appears that there are two different datasets - HumanML3D (22 joints) and The generated KiT version (21 joints). I'm interested in knowing if this dictionary categorizes all of the vocabulary found within these datasets.

— Reply to this email directly, view it on GitHub https://github.com/EricGuo5513/HumanML3D/issues/30#issuecomment-1403662087, or unsubscribe https://github.com/notifications/unsubscribe-auth/AKRYNB54HQZ5UOEUSOJV6OTWUEWFDANCNFSM6AAAAAAUFSBXRI . You are receiving this because you commented.Message ID: @.***>