Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2409.14128 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912133552275456 |
|---|---|
| author | Bernabeu-Perez, Pablo Lopez-Cuena, Enrique Garcia-Gasulla, Dario |
| author_facet | Bernabeu-Perez, Pablo Lopez-Cuena, Enrique Garcia-Gasulla, Dario |
| contents | The continued release of increasingly realistic image generation models creates a demand for synthetic image detectors. To build effective detectors we must first understand how factors like data source diversity, training methodologies and image alterations affect their generalization capabilities. This work conducts a systematic analysis and uses its insights to develop practical guidelines for training robust synthetic image detectors. Model generalization capabilities are evaluated across different setups (e.g. scale, sources, transformations) including real-world deployment conditions. Through an extensive benchmarking of state-of-the-art detectors across diverse and recent datasets, we show that while current approaches excel in specific scenarios, no single detector achieves universal effectiveness. Critical flaws are identified in detectors, and workarounds are proposed to enable the deployment of real-world detector applications enhancing accuracy, reliability and robustness beyond the limitations of current systems. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2409_14128 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Present and Future Generalization of Synthetic Image Detectors Bernabeu-Perez, Pablo Lopez-Cuena, Enrique Garcia-Gasulla, Dario Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning The continued release of increasingly realistic image generation models creates a demand for synthetic image detectors. To build effective detectors we must first understand how factors like data source diversity, training methodologies and image alterations affect their generalization capabilities. This work conducts a systematic analysis and uses its insights to develop practical guidelines for training robust synthetic image detectors. Model generalization capabilities are evaluated across different setups (e.g. scale, sources, transformations) including real-world deployment conditions. Through an extensive benchmarking of state-of-the-art detectors across diverse and recent datasets, we show that while current approaches excel in specific scenarios, no single detector achieves universal effectiveness. Critical flaws are identified in detectors, and workarounds are proposed to enable the deployment of real-world detector applications enhancing accuracy, reliability and robustness beyond the limitations of current systems. |
| title | Present and Future Generalization of Synthetic Image Detectors |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2409.14128 |