About affordance estimation

[kokuratkt]

Hello, thank you for your research.
When I generated motions using the same text as on the project page with the pretrained model, the affordance estimation didn’t seem to work properly — the way the objects are held looks unnatural.
Could you please tell me which parameters you used during generation and which npy files correspond to each object?
The following video shows the actual generation.

https://github.com/user-attachments/assets/8d6faf45-1f64-4701-9011-a7efc67c1b67
https://github.com/user-attachments/assets/c799f2c8-c23e-4e16-9574-f96105bf6338
https://github.com/user-attachments/assets/a9210a4b-4bd6-485e-a1c4-00777f9a4d2a