Deressa Wodajo Deressa

Deressa Wodajo Deressa, Hannes Mareen, Peter Lambert et al.

Deepfakes have raised significant concerns due to their potential to spread false information and compromise digital media integrity. Current deepfake detection models often struggle to generalize across a diverse range of deepfake generation techniques and video content. In this work, we propose a Generative Convolutional Vision Transformer (GenConViT) for deepfake video detection. Our model combines ConvNeXt and Swin Transformer models for feature extraction, and it utilizes Autoencoder and Variational Autoencoder to learn from the latent data distribution. By learning from the visual artifacts and latent data distribution, GenConViT achieves improved performance in detecting a wide range of deepfake videos. The model is trained and evaluated on DFDC, FF++, TM, DeepfakeTIMIT, and Celeb-DF (v$2$) datasets. The proposed GenConViT model demonstrates strong performance in deepfake video detection, achieving high accuracy across the tested datasets. While our model shows promising results in deepfake video detection by leveraging visual and latent features, we demonstrate that further work is needed to improve its generalizability, i.e., when encountering out-of-distribution data. Our model provides an effective solution for identifying a wide range of fake videos while preserving media integrity. The open-source code for GenConViT is available at https://github.com/erprogs/GenConViT.

Deressa Wodajo Deressa, Hannes Mareen, Peter Lambert et al.

We present Generative Anchored Fields (GAF), a generative model that learns independent endpoint predictors, $J$ (noise) and $K$ (data), from any point on a linear bridge. Unlike existing approaches that use a single trajectory or score predictor, GAF is trained to recover the bridge endpoints directly via coordinate learning. The velocity field $v=K-J$ emerges from their time-conditioned disagreement. This factorization enables \textit{Transport Algebra}: algebraic operations on multiple $J/K$ heads for compositional control. With class-specific $K_n$ heads, GAF defines directed transport maps between a shared base noise distribution and multiple data domains, allowing controllable interpolation, multi-class composition, and semantic editing. This is achieved either directly on the predicted data coordinates ($K$) using Iterative Endpoint Refinement (IER), a novel sampler that achieves high-quality generation in $5-8$ steps, or on the emergent velocity field ($v$). We achieve strong sample quality (FID 7.51 on ImageNet $256\times256$ and $7.27$ on CelebA-HQ $256\times 256$, without classifier-free guidance) while treating compositional generation as an architectural primitive. Code available at https://github.com/IDLabMedia/GAF.