We compare video generated by our method, that uses intermediate rendering of 3D meshes as conditioning to baseline method that uses 2D keypoints as intermediate representation. The 2D keypoints are extrated from Mediapipe from ground-truth video. We also provide the corresponding ground-truth video for comparison.
|
Keypoint Maps 2D Baseline Input Mesh Ours Ground Truth Video Oliver |
|---|
|
Keypoint Maps 2D Baseline Input Mesh Ours Ground Truth Video Conan |
|
Keypoint Maps 2D Baseline Input Mesh Ours Ground Truth Video Seth |
|
Keypoint Maps 2D Baseline Input Mesh Ours Ground Truth Video Chemistry |