My hypothesis for how markerless motion capture is executed is via computer vision methods or use of point cloud generated via Lidar or laser scans.
seenumadhavan
For the right paper, here is a link to the project page: https://akanazawa.github.io/human_dynamics/, and the paper: https://arxiv.org/pdf/1812.01601.pdf. Very cool work! If I understood properly, they used ResNet to extract features followed by a temporal encoder, and then a separate "hallucinator" which lets them predict future frames of motion from a static image. BTW, Prof Kanazawa, who is first author on these, is a professor here.
My hypothesis for how markerless motion capture is executed is via computer vision methods or use of point cloud generated via Lidar or laser scans.
For the right paper, here is a link to the project page: https://akanazawa.github.io/human_dynamics/, and the paper: https://arxiv.org/pdf/1812.01601.pdf. Very cool work! If I understood properly, they used ResNet to extract features followed by a temporal encoder, and then a separate "hallucinator" which lets them predict future frames of motion from a static image. BTW, Prof Kanazawa, who is first author on these, is a professor here.