Why Are Conditional Generative Models Better Than Unconditional Ones?
This work addresses the challenge of improving unconditional generative models for image synthesis, offering a novel approach that bridges the performance gap with conditional models, though it is incremental in nature.
The paper tackles the problem of why conditional generative models outperform unconditional ones, proposing self-conditioned diffusion models (SCDM) that use clustered indices for conditioning, achieving a record-breaking FID of 3.94 on ImageNet 64x64 without labels and slightly better FID than conditional models on CIFAR10.
Extensive empirical evidence demonstrates that conditional generative models are easier to train and perform better than unconditional ones by exploiting the labels of data. So do score-based diffusion models. In this paper, we analyze the phenomenon formally and identify that the key of conditional learning is to partition the data properly. Inspired by the analyses, we propose self-conditioned diffusion models (SCDM), which is trained conditioned on indices clustered by the k-means algorithm on the features extracted by a model pre-trained in a self-supervised manner. SCDM significantly improves the unconditional model across various datasets and achieves a record-breaking FID of 3.94 on ImageNet 64x64 without labels. Besides, SCDM achieves a slightly better FID than the corresponding conditional model on CIFAR10.