Thanks for your great work!
As you did not provide the checkpoint for motionagformer-b for gt_2d input on h36m dataset.
I trained it on my device(3090 * 2), but i get a result more better than you showed in paper.....
In your paper, the mpjpe of gt_2d input got motionagformer-b model is 19.3?
So, regarding the scenario where ground truth (GT) points are used as input, do the results presented in your paper reflect the model's performance after full training convergence? @siarez @SoroushMehraban
Thanks for your great work!
As you did not provide the checkpoint for motionagformer-b for gt_2d input on h36m dataset.
I trained it on my device(3090 * 2), but i get a result more better than you showed in paper.....
In your paper, the mpjpe of gt_2d input got motionagformer-b model is 19.3?
So, regarding the scenario where ground truth (GT) points are used as input, do the results presented in your paper reflect the model's performance after full training convergence? @siarez @SoroushMehraban