MIPT, Moscow, Russia
In NIPS 2019
Our proposed pipeline.
Modern photo editing tools allow creating realistic manipulated images easily. While fake images can be quickly generated, learning models for their detection is challenging due to the high variety of tampering artifacts and the lack of large labeled datasets of manipulated images. In this paper, we propose a new framework for training of discriminative segmentation model via an adversarial process.
Fantastic Reality Dataset poster.
The dataset is divided into two splits: ‘Rough’ and ‘Realistic’. ‘Rough’ split contains 8k splices with obvious artifacts such aliasing at splice edges, light and color inconsistencies. We use the ‘Rough’ split to allow gradual learning of our retouching generator GR. The ‘Realistic’ split provides 8k splices that were retouched manually to be visually indistinguishable from the authentic background image. We use the ‘Realistic’ split to test our model and baselines.
For download "Fantastic Reality Dataset" dataset write to vl.kniaz@gosniias.ru
Created date: 2022-10-13 09:51:40
Last update: 2022-10-13 09:56:30