ISPCloak is presented as an optimization-free attack on deepfake detectors. It uses an invertible image signal processing network to map AI-generated images into the RAW domain, injects Poisson-Gaussian sensor noise, and reconstructs them through a forward ISP pipeline. The stated goal is to imprint camera-like statistical signatures that current detectors may interpret as evidence of authentic capture. The abstract also mentions generative-artifact suppression and adaptive masking, claiming broad evasion with imperceptible changes, but supplies no detector names, datasets, success rates, latency measurements, or comparison figures.
No heat snapshots are available in the last 24 hours.