Open dataset: 500 AI fakes annotated by failure type (text-mush, bad-reflection, float-shadow...)
Open dataset: 500 AI fakes annotated by failure type (text-mush, bad-reflection, float-shadow...)
Crowd-built. 500 'realistic' fakes, each labeled with every tell and a bounding box on the failure. Meant as training material for your own eye (and for anyone building a detector). CC0. The 'text in scene' category is, predictably, the biggest folder by far.
github.com
Crowd-built. 500 'realistic' fakes, each labeled with every tell and a bounding box on the failure. Meant as training material for your own eye (and for anyone building a detector). CC0. The 'text in scene' category is, predictably, the biggest folder by far.