Institution profile

Hefei University of Technology

Academic institutionasia · cn
Official website
Research library570linked papers
Opportunities0open roles
Selected work

Representative Papers

Learning to Synthesize Compatible Fashion Items Using Semantic Alignment and Collocation Classification: An Outfit Generation Framework

Sep 15, 2022IEEE Transactions on Neural Networks and Learning Systems

This work addresses the challenging problem of complete outfit generation conditioned on a single garment and target-region masks—a key task in fashion design automation. We propose OutfitGAN, an end-to-end generative framework that synthesizes compatible tops, bottoms, footwear, and accessories given an input garment and spatially localized masks. Methodologically, we introduce two novel components: (i) a Semantic Alignment Module (SAM) that models fine-grained cross-garment semantic correspondences, and (ii) a Compatibility Classification Module (CCM) that explicitly enforces style and semantic coherence. Our multi-stage GAN architecture integrates semantic segmentation guidance, feature-level alignment losses, compatibility-aware adversarial supervision, and mask-conditioned generation control. Evaluated on a large-scale dataset of 20,000 real-world outfits, OutfitGAN achieves state-of-the-art performance across image fidelity, perceptual realism, and outfit compatibility metrics. It enables high-fidelity, diverse, and interactive fashion editing.

13 citationsRead paper

FCBoost-Net: A Generative Network for Synthesizing Multiple Collocated Outfits via Fashion Compatibility Boosting

Oct 26, 2023ACM Multimedia

To address the limitations of single-output generation, insufficient diversity, and difficulty in jointly optimizing compatibility in fashion outfit generation, this paper proposes a multi-round iterative optimization framework. The method builds upon a pre-trained generative model and introduces the novel Fashion Compatibility Booster (FCB)—a boosting-inspired mechanism that employs a compatibility discriminator to guide multiple rounds of unpaired image translation. This enables joint optimization of visual realism, stylistic diversity, and cross-category compatibility without requiring paired training data. Given a query item, the framework generates multiple coherent, visually plausible, and stylistically diverse complete outfits. Experiments demonstrate that our approach achieves a 23.5% improvement in compatibility over state-of-the-art methods, maintains 91.3% diversity retention, and significantly outperforms existing approaches across all three key metrics: compatibility, diversity, and visual fidelity.

5 citationsRead paper

DNF: Dual-Layer Nested Fingerprinting for Large Language Model Intellectual Property Protection

Jan 13, 2026

This work addresses the challenge of effectively protecting intellectual property in black-box deployments of large language models, where existing watermarking approaches are vulnerable to filtering, leakage, and adaptive attacks. The authors propose a novel dual-layer nested fingerprinting technique that integrates domain-specific stylistic cues with implicit semantic triggers to construct a hierarchical backdoor mechanism. This approach enables highly stealthy fingerprint activation with low perplexity—without relying on rare tokens—while preserving model utility. Evaluated on Mistral-7B, LLaMA-3-8B-Instruct, and Falcon3-7B-Instruct, the method achieves 100% activation rates, maintains downstream task performance, and demonstrates strong robustness against fine-tuning, model merging, and detection-based attacks, significantly enhancing the practicality and security of model ownership verification in black-box settings.

2 citationsRead paper

Breaking Coordinate Overfitting: Geometry-Aware WiFi Sensing for Cross-Layout 3D Pose Estimation

Jan 18, 2026

Existing WiFi-based human pose estimation methods rely on camera-coordinate supervision, making them prone to overfitting specific device layouts and limiting their generalization. This work proposes PerceptAlign, a framework that aligns WiFi and visual spaces through a lightweight geometry-aware coordinate unification procedure requiring only two checkerboards and a few photographs. The calibrated transceiver positions are encoded into high-dimensional geometric embeddings and fused with channel state information (CSI) features to enable layout-agnostic 3D pose estimation. PerceptAlign introduces, for the first time, a geometry-conditioned learning mechanism that effectively disentangles human motion from device layout. Evaluated on the largest cross-domain WiFi pose dataset to date, the method reduces in-domain error by 12.3% and achieves over 60% reduction in cross-domain error.

1 citationsRead paper

BalDRO: A Distributionally Robust Optimization based Framework for Large Language Model Unlearning

Jan 14, 2026

This work addresses the challenge of uneven sample-level difficulty in the unlearning process of large language models, which often leads to insufficient removal of targeted knowledge or excessive forgetting of unrelated information. To tackle this issue, the paper introduces distributionally robust optimization (DRO) into unlearning for the first time, proposing a min–max optimization framework: the inner loop constructs the worst-case distribution over the hardest-to-forget samples, while the outer loop updates model parameters under this distribution to achieve balanced forgetting. Two efficient variants are developed—BalDRO-G, based on a discrete approximation from GroupDRO, and BalDRO-DV, leveraging continuous weighting via the Donsker–Varadhan dual formulation. Experiments on the TOFU and MUSE benchmarks demonstrate that the proposed approach significantly outperforms existing methods, achieving a superior trade-off between effective unlearning and preservation of model utility.

1 citationsRead paper
Recent publications

Latest Papers