PromptStyler: Prompt-driven Style Generation for Source-free Domain Generalization
Junhyeong Cho, Gilhyun Nam, Sungyeon Kim, Hunmin Yang, Suha Kwak
Abstract
In a joint vision-language space, a text feature (e.g., from "a photo of a dog") could effectively represent its relevant image features (e.g., from dog photos). Also, a recent study has demonstrated the cross-modal transferability phenomenon of this joint space. From these observations, we propose PromptStyler which simulates various distribution shifts in the joint space by synthesizing diverse styles via prompts without using any images to deal with source-free domain generalization. The proposed method learns to generate a variety of style features (from "a S <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">∗</inf> style of a") via learnable style word vectors for pseudo-words S <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">∗</inf> . To ensure that learned styles do not distort content information, we force style-content features (from "a S <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">∗</inf> style of a [class]") to be located nearby their corresponding content features (from "[class]") in the joint vision-language space. After learning style word vectors, we train a linear classifier using synthesized style-content features. PromptStyler achieves the state of the art on PACS, VLCS, OfficeHome and DomainNet, even though it does not require any images for training.