{"year":"2023","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","date_created":"2023-08-22T14:18:43Z","oa_version":"Preprint","oa":1,"status":"public","title":"A data augmentation perspective on diffusion models and retrieval","date_updated":"2023-09-13T08:51:56Z","month":"04","article_processing_charge":"No","external_id":{"arxiv":["2304.10253"]},"day":"20","publication":"arXiv","author":[{"full_name":"Burg, Max F.","first_name":"Max F.","last_name":"Burg"},{"last_name":"Wenzel","first_name":"Florian","full_name":"Wenzel, Florian"},{"last_name":"Zietlow","first_name":"Dominik","full_name":"Zietlow, Dominik"},{"full_name":"Horn, Max","last_name":"Horn","first_name":"Max"},{"first_name":"Osama","last_name":"Makansi","full_name":"Makansi, Osama"},{"full_name":"Locatello, Francesco","id":"26cfd52f-2483-11ee-8040-88983bcc06d4","first_name":"Francesco","last_name":"Locatello","orcid":"0000-0002-4850-0683"},{"last_name":"Russell","first_name":"Chris","full_name":"Russell, Chris"}],"doi":"10.48550/arXiv.2304.10253","type":"preprint","abstract":[{"lang":"eng","text":"Diffusion models excel at generating photorealistic images from text-queries. Naturally, many approaches have been proposed to use these generative abilities to augment training datasets for downstream tasks, such as classification. However, diffusion models are themselves trained on large noisily supervised, but nonetheless, annotated datasets. It is an open question whether the generalization capabilities of diffusion models beyond using the additional data of the pre-training process for augmentation lead to improved downstream performance. We perform a systematic evaluation of existing methods to generate images from diffusion models and study new extensions to assess their benefit for data augmentation. While we find that personalizing diffusion models towards the target data outperforms simpler prompting strategies, we also show that using the training data of the diffusion model alone, via a simple nearest neighbor retrieval procedure, leads to even stronger downstream performance. Overall, our study probes the limitations of diffusion models for data augmentation but also highlights its potential in generating new training data to improve performance on simple downstream vision tasks."}],"main_file_link":[{"url":"https://doi.org/10.48550/arXiv.2304.10253","open_access":"1"}],"_id":"14209","article_number":"2304.10253","language":[{"iso":"eng"}],"department":[{"_id":"FrLo"}],"extern":"1","citation":{"short":"M.F. Burg, F. Wenzel, D. Zietlow, M. Horn, O. Makansi, F. Locatello, C. Russell, ArXiv (n.d.).","ista":"Burg MF, Wenzel F, Zietlow D, Horn M, Makansi O, Locatello F, Russell C. A data augmentation perspective on diffusion models and retrieval. arXiv, 2304.10253.","chicago":"Burg, Max F., Florian Wenzel, Dominik Zietlow, Max Horn, Osama Makansi, Francesco Locatello, and Chris Russell. “A Data Augmentation Perspective on Diffusion Models and Retrieval.” <i>ArXiv</i>, n.d. <a href=\"https://doi.org/10.48550/arXiv.2304.10253\">https://doi.org/10.48550/arXiv.2304.10253</a>.","apa":"Burg, M. F., Wenzel, F., Zietlow, D., Horn, M., Makansi, O., Locatello, F., &#38; Russell, C. (n.d.). A data augmentation perspective on diffusion models and retrieval. <i>arXiv</i>. <a href=\"https://doi.org/10.48550/arXiv.2304.10253\">https://doi.org/10.48550/arXiv.2304.10253</a>","ama":"Burg MF, Wenzel F, Zietlow D, et al. A data augmentation perspective on diffusion models and retrieval. <i>arXiv</i>. doi:<a href=\"https://doi.org/10.48550/arXiv.2304.10253\">10.48550/arXiv.2304.10253</a>","ieee":"M. F. Burg <i>et al.</i>, “A data augmentation perspective on diffusion models and retrieval,” <i>arXiv</i>. .","mla":"Burg, Max F., et al. “A Data Augmentation Perspective on Diffusion Models and Retrieval.” <i>ArXiv</i>, 2304.10253, doi:<a href=\"https://doi.org/10.48550/arXiv.2304.10253\">10.48550/arXiv.2304.10253</a>."},"publication_status":"submitted","date_published":"2023-04-20T00:00:00Z"}