Beyond Detection: Multi-Scale Hidden-Code for Natural Image Deepfake Recovery and Factual Retrieval

TL;DR

This paper introduces a unified framework for recovering and retrieving tampered image contents using multi-scale hidden-code encoding, enhancing factual retrieval and restoration beyond traditional deepfake detection methods.

Contribution

It presents a novel hidden-code recovery approach with multi-scale vector quantization and Transformer modules, along with a new ImageNet-S benchmark for evaluating image retrieval and restoration.

Findings

Demonstrates effective retrieval and reconstruction on ImageNet-S

Compatible with various watermarking pipelines

Establishes a foundation for general-purpose image recovery

Abstract

Recent advances in image authenticity have primarily focused on deepfake detection and localization, leaving recovery of tampered contents for factual retrieval relatively underexplored. We propose a unified hidden-code recovery framework that enables both retrieval and restoration from post-hoc and in-generation watermarking paradigms. Our method encodes semantic and perceptual information into a compact hidden-code representation, refined through multi-scale vector quantization, and enhances contextual reasoning via conditional Transformer modules. To enable systematic evaluation for natural images, we construct ImageNet-S, a benchmark that provides paired image-label factual retrieval tasks. Extensive experiments on ImageNet-S demonstrate that our method exhibits promising retrieval and reconstruction performance while remaining fully compatible with diverse watermarking pipelines. This framework establishes a foundation for general-purpose image recovery beyond detection and localization.