Formatting the Visible. In the Factory of Photo Datasets (2005‑2021)
A core element of the AI algorithms applied to images, photo datasets illustrate the central part that photography plays in these machine learning systems. But what do these vast sets of images reveal? And what operations are involved? The visual regimes that were observed in twenty-five datasets ev...
Enregistré dans:
| 發表在: | URI:https://journals.openedition.org/transbordeur, |
|---|---|
| 主要作者: | |
| 格式: | Article ou chapitre numérique |
| 語言: | Anglais |
| 出版: |
Transbordeur
2025
|
| 主題: | |
| 在線閱讀: | Accès Université d'Orléans et IFPM Accès Université d'Orléans et IFPM |
| 總結: | A core element of the AI algorithms applied to images, photo datasets illustrate the central part that photography plays in these machine learning systems. But what do these vast sets of images reveal? And what operations are involved? The visual regimes that were observed in twenty-five datasets evolved between 2005 and 2021: limited sets of standardized, descriptive images made way for massive online collections that are automated and uncurated. The practices and protocols used for selecting and annotating and the visual processing performed on the photographs are meant to neutralize bias and sample a visible world that is fragmented and decontextualized. More than a reflection of reality, these datasets thus express a reconstruction of it in the service of an increasingly complex set of algorithmic problems. Photography has become a raw material that is exploited and crafted in a process of industrial extraction. |
|---|