Shinichi Tanaka

me too. Furthermore, I cant reproduce the real score of m2t on this MotionGPT paper and MotionGPT-2 paper. In these papers, R-Precision, MM Dist. are following:  However, when I...

@lucidrains Thank you for your reply!!! I'm currently working on implementing adaptation for motion data. So, without questioning previous research, I assumed that positional embeddings should be applied to motion...

@lucidrains Thank you for sharing your knowledge and the paper! Regarding motion data: it has a shape of (num_frames, 263), where each frame is treated as one token on transformer-based...

@lucidrains wow... Thank you so much!!! You've made it possible to set arbitrary Positional Encoding for various modalities while still applying only Rotary Embedding to text. This is greatly appreciated!!!

I am implementing diffusion model based transfusion, not your flow matching. It will take some time for me to produce results that are free of implementation misses with my implementation...