CompLift accepts or rejects generated samples by computing lift scores from conditional and unconditional denoising errors, improving compositional alignment without retraining.
K., Wang, Y .-X., and Hebert, M
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Improving Compositional Generation with Diffusion Models Using Lift Scores
CompLift accepts or rejects generated samples by computing lift scores from conditional and unconditional denoising errors, improving compositional alignment without retraining.