Search papers, labs, and topics across Lattice.
University, Kiel University, Mickiewicz University
2
0
4
Predicting how well synthetic data will transfer to real-world tasks doesn't come down to just appearance or geometry, but their surprising interplay.
By isolating the subject from the background, MaSC achieves near-perfect accuracy (AUC=0.992) in distinguishing same-subject from cross-subject image pairs, exposing the limitations of existing global metrics for evaluating concept-driven generation.