This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
In this paper, we propose a pretrained reversible generation (PRG) framework that extracts unsupervised representations by reversing the generation process of pretrained continuous generative models, which is not fully explored in discriminative tasks. PRG leverages the high capacity of pretrained generative models to build a robust and generalizable feature extractor, while allowing flexible selection of feature hierarchies tailored to specific subtasks. It outperforms existing approaches on various benchmarks, achieving a top-1 accuracy of 78% on ImageNet at 64x64 resolution, and is the state-of-the-art among generative model-based methods. In addition, we verify the effectiveness of our approach through various ablation studies and out-of-distribution evaluations. The source code is available at https://github.com/opendilab/PRG .