This paper introduces a novel framework for multi-concept customization in text-to-image diffusion models . At its core is a Multiple Prompt Adapter (MP-Adapter) capable of processing multiple image prompts in parallel, extracting features from target concepts and projecting them into the same latent space as the text prompt. This enables simultaneous handling of multiple concepts using just one reference image per concept. To address challenges in fusing multiple concepts with complex interactions, we propose a Region-based Denoising Framework (RDF) that dynamically generates concept-specific regions of interest during inference, allowing spatially decoupled injection of concept features. By integrating the MP-Adapter and RDF, our end-to-end pipeline enables multi-concept customization with intricate occlusions and interactions while preserving concept identities. This approach surpasses current methods by resolving concept conflicts, identity degradation, and occlusion issues, allowing flexible customization without concept-specific retraining. Both qualitative and quantitative evaluations demonstrate that our framework outperforms state-of-the-art approaches in multi-concept customization tasks, while ablation studies validate the effectiveness of each proposed component. This work significantly advances text-to-image generation capabilities for complex, user-defined concept combinations. Code and models will be released at https://github.com/baojudezeze/RMP-Adapter .

Discrete logarithm problem (DLP) is one of the fundamental hard problems used in cryptography. For 1 ≤ k ≤ n , solving the k -out-of- n DLP instances is an important problem emerging in certain scenarios in public-key cryptography. Ying and Kunihiro (ACNS 2017) pioneered in studying k -out-of- n instance solutions of DLP, which is a generalized version of multiple DLP. By reducing the multiple DLP to the generalized version, they established lower bounds on the computational complexity of k -out-of- n DLP for different parameter values of k . In this paper, we further reduce the reduction complexity presented in Ying and Kunihiro's work and increase the range of k and n for the tight lower bound of k -out-of- n DLP in the generic group model, which has applications in related cryptographic schemes. To achieve the goal, the key technique is to utilize a variant of fast multipoint evaluation. We divide the discussion into two cases. In the special case when n divides p − 1 , by leveraging Number Theory Transform (NTT) technique, we expand k and n to a larger range. In the general case, by using a variant of fast multipoint evaluation, we increase k and n to a moderately larger range. • The complexity of reducing the k-MDL problem to the (k, n)-GMDL problem has been effectively lowered. • The range of parameters when (k, n)-GMDL problem achieves the tight complexity bound is expanded. • The main techniques are a variant of Number Theory Transform and a variant of fast multipoint evaluation.

To enhance the recognition accuracy of deep learning models in the domain of camera source recognition, we have developed an adaptive attention dense network structure and introduced an adaptive weighted attention method. The proposed network structure is composed of four key components: a preprocessing module, a dense connection module, an attention module, and a regularization module. Central to the attention mechanism, the adaptive weighted attention method utilizes an adaptive weight factor to optimize parameters dynamically, enabling the model to adapt to varying data characteristics. This approach enhances the model’s feature learning and representation capabilities. Extensive comparative and ablation experiments were conducted on two benchmark datasets. In the comparative experiments, our network was evaluated against three well-established networks. The experimental results demonstrated that our network’s recognition performance surpassed that of the other networks by at least 5.6% on one dataset and 10% on the other. And the ablation studies revealed that the proposed method improved recognition performance by at least 2.6% and 5.2% compared to other ablation methods across the two datasets.

The human pose transfer task aims to generate synthetic person images that preserve the style of reference images while accurately aligning them with the desired target pose. However, existing methods based on generative adversarial networks (GANs) struggle to produce realistic details and often face spatial misalignment issues. On the other hand, methods relying on denoising diffusion models require a large number of model parameters, resulting in slower convergence rates. To address these challenges, we propose a self-calibration flow-guided module (SCFM) to establish precise spatial correspondence between reference images and target poses. This module facilitates the denoising diffusion model in predicting the noise at each denoising step more effectively. Additionally, we introduce a multi-scale feature fusing module (MSFF) that enhances the denoising U-Net architecture through a cross-attention mechanism, achieving better performance with a reduced parameter count. Our proposed model outperforms state-of-the-art methods on the DeepFashion and Market-1501 datasets in terms of both the quantity and quality of the synthesized images. Our code is publicly available at https://github.com/zylwithxy/SCFM-guided-DDPM.