Tag archive
When making a fist, the thumb passes through the index finger. Four different pushing directions were tried, but all resulted in a thumbs up. The actual fist has the thumb on top of the index finger. Additionally, five failures where the guaranteed values from the search were ruined by smoothing.
The wrist twist was measured to be around 6-8 degrees, and the collapse was due to bending, not the pose, which was correct, but the skinning was incorrect, also discussing rig mobility, manufacturing failures, and WSL and Blender issues.
Limiting the leg length to 99% from the waist to the ankle resulted in a 16-18° bend in the knee, even in a standing frame. The closer the knee is to being straight, the more the angle is affected by the square root of the length deficiency. From ground detection to hip height.
The arm's interference correction oscillated 17cm forward and 15cm backward despite being stationary, due to the equilibrium of penetration depths at 74mm and 73mm. The escape direction is determined using candidate solutions and dynamic planning for the entire clip.
The angle difference between the head and chest was exactly 0.000° across the entire animation. After rebuilding the downstream process three times, I realized that the two parts I thought I was measuring separately were actually generated from the same torso orientation. A story about how the measurement process itself was flawed.
A pipeline to convert a single fixed-camera live-action video into VRM character animations. While the estimation itself is accurate, the character suffers from feet clipping through the floor and arms merging into the body. This is the first part of a six-part series on measuring and fixing these artifacts.
Recently I'm working on handtracking ARs. To align 3D models to hands, I had to match coordinates of...