Adaptive Identity Anchoring: Closed-Loop Keyframe Placement for Synthetic Paired Supervision in Video Face Swapping
The paper proposes adding identity anchors where a generated face swap drifts most, instead of only at the first and last frames.
It describes a closed-loop method that scores generated frames against the reference identity, then inserts a swapped anchor at the weakest frame until quality passes a threshold or the budget runs out. The same loop can reject bad synthetic pairs before they train a student model. It also adds texture restoration aimed at reducing the over-smoothed “beauty-filter” look by borrowing spectrum and micro-texture cues from the real footage. ArXiv · AI/CL/LG's note
It describes a closed-loop method that scores generated frames against the reference identity, then inserts a swapped anchor at the weakest frame until quality passes a threshold or the budget runs out. The same loop can reject bad synthetic pairs before they train a student model. It also adds texture restoration aimed at reducing the over-smoothed “beauty-filter” look by borrowing spectrum and micro-texture cues from the real footage. ArXiv · AI/CL/LG's note
score 4