Search papers, labs, and topics across Lattice.
2
0
4
7
By forcing a model to argue against its own forgery hypothesis using physical laws, FakeVLM-R1 achieves SOTA synthetic image detection and avoids the over-rejection bias of existing methods.
Freeing adversarial perturbations from pixel-level constraints by injecting them into the latent space of diffusion models unlocks significantly more robust defense against facial deepfakes.
