Closed ethan-hibikirun closed 10 months ago
saw it here https://paperswithcode.com/paper/motionbert-unified-pretraining-for-human#code but can't find the model. Is this 3D Pose (H36M-SH, ft)? but why are the metrics different
I think it is the result of using ground-truth 2D keypoint as input. You can train it by setting gt_2d as true in the training configs.
gt_2d
saw it here https://paperswithcode.com/paper/motionbert-unified-pretraining-for-human#code but can't find the model. Is this 3D Pose (H36M-SH, ft)? but why are the metrics different