LogicalDefender: Discovering, Extracting, and Utilizing Common-Sense Knowledge
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909139761889280 |
|---|---|
| author | Liu, Yuhe Kang, Mengxue Qin, Zengchang Chu, Xiangxiang |
| author_facet | Liu, Yuhe Kang, Mengxue Qin, Zengchang Chu, Xiangxiang |
| contents | Large text-to-image models have achieved astonishing performance in synthesizing diverse and high-quality images guided by texts. With detail-oriented conditioning control, even finer-grained spatial control can be achieved. However, some generated images still appear unreasonable, even with plentiful object features and a harmonious style. In this paper, we delve into the underlying causes and find that deep-level logical information, serving as common-sense knowledge, plays a significant role in understanding and processing images. Nonetheless, almost all models have neglected the importance of logical relations in images, resulting in poor performance in this aspect. Following this observation, we propose LogicalDefender, which combines images with the logical knowledge already summarized by humans in text. This encourages models to learn logical knowledge faster and better, and concurrently, extracts the widely applicable logical knowledge from both images and human knowledge. Experiments show that our model has achieved better logical performance, and the extracted logical knowledge can be effectively applied to other scenarios. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_11570 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | LogicalDefender: Discovering, Extracting, and Utilizing Common-Sense Knowledge Liu, Yuhe Kang, Mengxue Qin, Zengchang Chu, Xiangxiang Computer Vision and Pattern Recognition Large text-to-image models have achieved astonishing performance in synthesizing diverse and high-quality images guided by texts. With detail-oriented conditioning control, even finer-grained spatial control can be achieved. However, some generated images still appear unreasonable, even with plentiful object features and a harmonious style. In this paper, we delve into the underlying causes and find that deep-level logical information, serving as common-sense knowledge, plays a significant role in understanding and processing images. Nonetheless, almost all models have neglected the importance of logical relations in images, resulting in poor performance in this aspect. Following this observation, we propose LogicalDefender, which combines images with the logical knowledge already summarized by humans in text. This encourages models to learn logical knowledge faster and better, and concurrently, extracts the widely applicable logical knowledge from both images and human knowledge. Experiments show that our model has achieved better logical performance, and the extracted logical knowledge can be effectively applied to other scenarios. |
| title | LogicalDefender: Discovering, Extracting, and Utilizing Common-Sense Knowledge |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2403.11570 |