search

yzhq97 / transmomo.pytorch

This is the official PyTorch implementation of the CVPR 2020 paper "TransMoMo: Invariance-Driven Unsupervised Video Motion Retargeting".
https://yzhq97.github.io/transmomo/
391 stars 54 forks source link

Which point is the center for limb scaling? #3

Closed ailias closed 4 years ago

ailias commented 4 years ago

Q1: The result is amazing no matter what people is small or large. But I don't find the code you choose which point as the center for scaling structure, so all limbs will shifting together, it is not reasonable. And if we scaling every target's skeleton frame by frame independently, whether the generated result will jittering?

Q2: For all target video, do we just need one common model for generating results?

Walter0807 commented 4 years ago

A1: Limb-scaling process starts from the root joint (pelvis). The length of a limb is extended/shortened by the same ratio across all frames, so it does not introduce ambiguity. Please also note that limb scaling is only applied at the training stage as a perturbation, not at the inference stage.

A2: The results we show are generated by separate models for each target. "One common model" is also possible, depending on the specific skeleton-to-video rendering technique.