Walter0807 / MotionBERT

[ICCV 2023] PyTorch Implementation of "MotionBERT: A Unified Perspective on Learning Human Motion Representations"
Apache License 2.0
1.06k stars 131 forks source link

How to finetune action recognition? #86

Open hotelbread opened 1 year ago

hotelbread commented 1 year ago

Hi. first of all, I appreciate your working. It is awesome!! I always think that your idea, there are many task but difference format and units. I wondered if combining them could lead to better performance.

Now I wanna finetuning with my own datasets which has own classes. Due to my limited knowledge, I struggled to apply the concepts effectively. I found other question #23 , but it is also lack of information to me.

and then I also wanna finetuning with my own dataset for 3Dpose, but It would be next step, after when I success to finetuining action-recognition.

your action recognition is trained by NTU-RGBD, then do I need same format dataset to finetuing?

I understand that this question is relatively basic, so I am aware that answering it might be a bit bothersome. However, it must be helpful for beginner.

Thanks! seotae89@gmail.com

hellos0032 commented 3 months ago

Hi. I also wondering about combination performance. Did you get the result of it ?

Thank you!