The paper introduces ViT‑HarDis, a method that conditions attention head disentanglement on sample difficulty in Vision Transformers for face recognition. By applying stronger head decorrelation to hard samples—those with high cosine similarity to negative class centers—the approach encourages specialization where identity ambiguity is greatest. Experiments on WebFace4M and MS1MV3 show that ViT‑HarDis improves over baseline and uniform disentanglement, achieving state‑of‑the‑art results on IJB‑B, IJB‑C, and TinyFace without architectural changes or extra inference cost.
arXiv:2609.01014v1 Announce Type: new Abstract: Low-quality face recognition (LQFR) remains challenging due to the difficulty of matching degraded query (probe) images against low-quality (LQ) enroll...
Low-quality face recognition (LQFR) remains challenging due to the difficulty of matching degraded query (probe) images against low-quality (LQ) enrollment (gallery) imagery and the scarcity of traini...
arXiv:2605.07191v2 Announce Type: replace-cross Abstract: A recent work shows that Attention Transfer, which transfers only the attention patterns from a pre-trained teacher Vision Transformer (ViT)...
arXiv:2607.09086v2 Announce Type: replace Abstract: We present Subtoken Vision Transformer (SubViT), a selective image tokenization method for fine-grained visual recognition. Standard Vision Transfo...
The paper introduces an importance‑scoring metric for multi‑head transformer attention heads applied to tabular data, a domain where transformers have been less studied. Experiments on 40 diverse tabular datasets show that removing heads with the lowest importance scores has minimal impact on performance, while removing the most important head first causes the largest drop. The study finds that important heads are distributed across layers and vary significantly across different tabular schemas, suggesting that the proposed score can help reduce redundancy and improve transformer efficiency.
arXiv:2607. 26993v1 Announce Type: new Abstract: Face presentation attack detection (PAD) remains challenging under cross-dataset evaluation, where domain shift degrades models trained on a single dataset.
GM-GReFEL is a geometry‑guided spatiotemporal state‑space network for dynamic facial expression recognition that replaces a Vision Transformer with a Modulated GroupMamba backbone, reducing self‑attention cost to linear time. It incorporates a reliability module that mixes classifier outputs with geometric anchor votes based on prediction entropy, and is trained via a Triple Loss after static‑to‑dynamic weight transfer. In five‑fold cross‑validation it achieves competitive UAR/WAR scores on DFEW, MAFW, and FERV39k, with modest temporal modeling benefits and latency comparable to ViT‑B/16.
arXiv:2608.23921v1 Announce Type: new Abstract: Recent Vision-Language Models encode high-resolution images into long visual token sequences, incurring prohibitive prefill costs. To compress them, ex...
CoViT introduces a self‑supervised framework that enhances Vision Transformers with instance‑aware representations by leveraging geometry‑guided contrastive learning. It refines attention maps to generate instance masks and constructs triplets that mine the hardest intra‑ and inter‑instance examples, driving a contrastive loss that reduces intra‑instance variance while increasing inter‑instance margins. The method yields consistent AP gains of over 2 points on instance‑level tasks without requiring extra decoders or labels.
The paper audits the MS1MV3 face‑recognition training dataset and identifies five defect categories—exact/near duplicates, split identities, crop errors, multiple‑face ambiguity, and invalid content. Using IResNet‑50 and ArcFace, the authors show that correcting these defects, especially duplicate removal, can maintain or improve recognition performance while reducing training effort. The study highlights that dataset quality significantly influences large‑scale face‑recognition outcomes.
LearnPruner rethinks token pruning in vision‑language models by addressing shortcomings of attention mechanisms in vision encoders and large language models. The method uses a two‑stage approach: first a learnable pruning module removes redundant vision tokens after the vision encoder, then only task‑relevant tokens are kept in the LLM’s middle layer. Experiments show that LearnPruner retains about 95% of performance while using only 5.5% of vision tokens, achieving a 3.2× speed‑up.
The paper introduces Dual Quality Margin Learning for Face Recognition (DQM‑Face), a framework that combines magnitude‑based and semantic quality estimation to refine attraction and repulsion dynamics during training. By integrating squeeze‑and‑excitation semantic attention with dual margins, the method enhances intra‑class compactness and inter‑class separation, yielding a more discriminative feature geometry. Experiments on challenging benchmarks show that DQM‑Face outperforms state‑of‑the‑art face recognition models and that the learned quality signal aligns well with recognition objectives.