We present CSWin Transformer, an efficient and effective Transformer-based backbone for general-purpose vision tasks. A challenging issue in Transformer design is that global self-attention very expensive to compute whereas local often limits the field of interactions each token. To address this issue, we develop Cross-Shaped Window mechanism computing horizontal vertical stripes parallel form a cross-shaped window, with stripe obtained by splitting input feature into equal width. provide...

Recently, more and attention is paid to reversible data hiding (RDH) in encrypted images, since it maintains the excellent property that original cover can be losslessly recovered after embedded extracted while protecting image content's confidentiality. All previous methods embed by reversibly vacating room from which may subject some errors on extraction and/or restoration. In this paper, we propose a novel method reserving before encryption with traditional RDH algorithm, thus easy for...

Face forgery by deepfake is widely spread over the internet and has raised severe societal concerns. Recently, how to detect such contents become a hot research topic many detection methods have been proposed. Most of them model as vanilla binary classification problem, i.e, first use backbone network extract global feature then feed it into classifier (real/fake). But since difference between real fake images in this task often subtle local, we argue solution not optimal. In paper, instead...

Prediction-error expansion (PEE) is the most successful reversible data hiding (RDH) technique, and existing PEE-based RDH methods are mainly based on modification of one- or two-dimensional prediction-error histogram (PEH). The PEH-based perform generally better than those one-dimensional PEH; however, their performance still unsatisfactory since PEH manner fixed independent image content. In this paper, we propose a new method PEE for multiple histograms. Unlike previous methods, consider...

The remarkable success in face forgery techniques has received considerable attention computer vision due to security concerns. We observe that up-sampling is a necessary step of most techniques, and cumulative will result obvious changes the frequency domain, especially phase spectrum. According property natural images, spectrum preserves abundant components provide extra information complement loss amplitude To this end, we present novel Spatial-Phase Shallow Learning (SPSL) method, which...

In this paper, based on two-dimensional difference- histogram modification, a novel reversible data hiding (RDH) scheme is proposed by using difference-pair-mapping (DPM). First, considering each pixel-pair and its context, sequence consisting of pairs difference values computed. Then, difference-histogram generated counting the frequency resulting difference-pairs. Finally, embedding implemented according to specifically designed DPM. Here, DPM an injective mapping defined It natural...

State-of-the-art schemes for reversible data hiding (RDH) usually consist of two steps: first construct a host sequence with sharp histogram via prediction errors, and then embed messages by modifying the methods, such as difference expansion shift. In this paper, we focus on second stage, propose modification method RDH, which embeds message recursively utilizing decompression compression processes an entropy coder. We prove that, independent identically distributed (i.i.d.) gray-scale...

In this work we propose Identity Consistency Transformer, a novel face forgery detection method that focuses on high-level semantics, specifically identity information, and detecting suspect by finding inconsistency in inner outer regions. The Transformer incorporates consistency loss for determination. We show exhibits superior generalization ability not only across different datasets but also various types of image degradation forms found real-world applications including deepfake videos....

This paper presents a simple yet effective framework MaskCLIP, which incorporates newly proposed masked self-distillation into contrastive language-image pretraining. The core idea of is to distill representation from full image the predicted image. Such incorporation enjoys two vital benefits. First, targets local patch learning, complementary vision-language focusing on text-related representation. Second, also consistent with perspective training objective as both utilize visual encoder...

Face swapping has drawn a lot of attention for its compelling performance. However, current deepfake methods suffer the effects obscure workflow and poor To solve these problems, we present DeepFaceLab, dominant framework practical face-swapping. It provides necessary tools as well an easy-to-use way to conduct high-quality also offers flexible loose coupling structure people who need strengthen their pipeline with other features without writing complicated boilerplate code. We detail...

With the popularity of outsourcing data to cloud, it is vital protect privacy and enable cloud server easily manage at same time. Under such demands, reversible hiding in encrypted images (RDH-EI) attracts more researchers' attention. In this paper, we propose a novel framework for RDH-EI based on image transformation (RIT). Different from all previous encryption-based frameworks, which ciphertexts may attract notation curious RIT-based allows user transform content original into another...

This paper proposes a novel screen-shooting resilient watermarking scheme, which means that if the watermarked image is displayed on screen and information captured by camera, we can still extract watermark message from photo. To realize such demands, analyzed special distortions caused process, including lens distortion, light source moiré distortion. resist geometric deformation proposed an intensity-based scale-invariant feature transform (I-SIFT) algorithm accurately locate embedding...

Neural networks are vulnerable to adversarial examples, which poses a threat their application in security sensitive systems. We propose Denoiser and UPsampler Network (DUP-Net) structure as defenses for 3D point cloud classification, where the two modules reconstruct surface smoothness by dropping or adding points. In this paper, statistical outlier removal (SOR) data-driven upsampling network considered denoiser upsampler respectively. Compared with baseline defenses, DUP-Net has three...

Deep Neural Networks (DNNs) have recently led to significant improvements in many fields. However, DNNs are vulnerable adversarial examples which samples with imperceptible perturbations while dramatically misleading the DNNs. Moreover, can be used perform an attack on various kinds of DNN based systems, even if adversary has no access underlying model. Many defense methods been proposed, such as obfuscating gradients networks or detecting examples. However it is proved out that these not...

While most techniques of reversible data hiding in encrypted images (RDH-EI) are developed for uncompressed images, this paper provides a separable protocol JPEG bitstreams. We first propose encryption algorithm, which enciphers an image to smaller size and keeps the format compliant decoder. After content owner uploads bitstream remote server, hider embeds additional message into copy without changing size. On recipient side, original can be reconstructed losslessly using iterative recovery...

Reversible data hiding in encrypted domain (RDH-ED) has greatly attracted researchers as the original content can be losslessly reconstructed after embedded are extracted, while owner's privacy remains protected. Most of existing RDH-ED algorithms designed for grayscale/color images, which cannot directly applied to other carriers, such three-dimensional (3D) meshes. With rapid development 3D related applications, models have been widely used on Internet, motivated us design a reliable...

Pairwise prediction-error expansion (pairwise PEE) is a recent technique for the high-dimensional reversible data hiding. However, in absence of adaptive embedding, its potential has not been fully exploited. In this paper, we propose pixel pairing (APP) and mapping selection enhancement pairwise PEE. Our motivation twofold: building sharper 2D histogram designing effective it. APP, consider to increase similarity between pixels pair, by excluding rough from only putting smooth into pairs....

Recent advances on adaptive steganography imply that the security of can be improved by exploiting mutual impact modifications between adjacent cover elements, such as pixels images, which is called a nonadditive distortion model. In this paper, we propose framework for defining joint pixel blocks. To reduce complexity minimizing distortion, design coding method to decompose (abbreviated DeJoin) into individual pixels; thus, message efficiently embedded with syndrome-trellis codes. We prove...

Despite the tremendous success, deep neural networks are exposed to serious IP infringement risks. Given a target model, if attacker knows its full information, it can be easily stolen by fine-tuning. Even only output is accessible, surrogate model trained through student-teacher learning generating many input-output training pairs. Therefore, protection important and necessary. However, still seriously under-researched. In this work, we propose new watermarking framework for protecting...

Deep neural network based steganalysis has developed rapidly in recent years, which poses a challenge to the security of steganography. However, there is no steganography method that can effectively resist networks for at present. In this paper, we propose new strategy constructs enhanced covers against with technique adversarial examples. The and their corresponding stegos are most likely be judged as by networks. Besides, use both deep high-dimensional feature classifiers evaluate...