Attributed Synthetic Data Generation for Zero-shot Domain-specific Image Classification
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866916676088365056 |
|---|---|
| author | Wang, Shijian Song, Linxin Shimizu, Ryotaro Goto, Masayuki Wu, Hanqian |
| author_facet | Wang, Shijian Song, Linxin Shimizu, Ryotaro Goto, Masayuki Wu, Hanqian |
| contents | Zero-shot domain-specific image classification is challenging in classifying real images without ground-truth in-domain training examples. Recent research involved knowledge from texts with a text-to-image model to generate in-domain training images in zero-shot scenarios. However, existing methods heavily rely on simple prompt strategies, limiting the diversity of synthetic training images, thus leading to inferior performance compared to real images. In this paper, we propose AttrSyn, which leverages large language models to generate attributed prompts. These prompts allow for the generation of more diverse attributed synthetic images. Experiments for zero-shot domain-specific image classification on two fine-grained datasets show that training with synthetic images generated by AttrSyn significantly outperforms CLIP's zero-shot classification under most situations and consistently surpasses simple prompt strategies. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_04510 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Attributed Synthetic Data Generation for Zero-shot Domain-specific Image Classification Wang, Shijian Song, Linxin Shimizu, Ryotaro Goto, Masayuki Wu, Hanqian Computer Vision and Pattern Recognition Zero-shot domain-specific image classification is challenging in classifying real images without ground-truth in-domain training examples. Recent research involved knowledge from texts with a text-to-image model to generate in-domain training images in zero-shot scenarios. However, existing methods heavily rely on simple prompt strategies, limiting the diversity of synthetic training images, thus leading to inferior performance compared to real images. In this paper, we propose AttrSyn, which leverages large language models to generate attributed prompts. These prompts allow for the generation of more diverse attributed synthetic images. Experiments for zero-shot domain-specific image classification on two fine-grained datasets show that training with synthetic images generated by AttrSyn significantly outperforms CLIP's zero-shot classification under most situations and consistently surpasses simple prompt strategies. |
| title | Attributed Synthetic Data Generation for Zero-shot Domain-specific Image Classification |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2504.04510 |