Search papers, labs, and topics across Lattice.
This paper introduces CI-Diff, a novel Counterfactual Inference-based Diffusion approach designed to enhance rare concept generation by overcoming the common knowledge bias inherent in diffusion models. By reformulating the classifier-free guidance mechanism, CI-Diff effectively decouples unusual attributes from their associated concepts, allowing for more accurate synthesis of images based on text prompts describing rare attributes. Extensive experiments on the RareBench benchmark demonstrate that CI-Diff outperforms existing state-of-the-art diffusion models in generating images that align with rare concepts.
CI-Diff successfully decouples unusual attributes from common associations, enabling accurate image synthesis for rare concepts that previous models struggled to render.
Rare concept generation focuses on synthesizing customized images conditioned on text prompts that describe objects with unusual attributes. Previous works failed to align the generated images with rare concepts, resulting in incorrect attribute rendering or inconsistent composition of concepts. Such failures, as we observed, stem from the inherent common knowledge bias in the training stage of diffusion models, where objects are strongly associated with their common attributes, making it difficult to break these associations when generating rare concepts. To address such challenges, in this paper, we propose a novel Counterfactual Inference-based Diffusion approach, dubbed CI-Diff. CI-Diff blocks the interference of the model's inherent common knowledge bias and utilizes the Natural Direct Effect to capture the independent influence of the text prompt of rare concepts on image generation so that decoupling the unusual attributes from the rare concepts. To this end, we reformulate the classifier-free guidance mechanism to highlight the atypical attributes. To the best of our knowledge, we are the first to introduce causal inference into the rare concept generation task. Extensive experiments on the RareBench benchmark validate the superiority of CI-Diff over state-of-the-art diffusion models. Our code can be accessed from https://github.com/200204jzy/CI-Diff.
