Zhihui Fu

Ruiguang Pei, Weiqing Sun, Zhihui Fu et al.

Although Large Vision Language Models (LVLMs) have demonstrated remarkable performance in image understanding tasks, their computational efficiency remains a significant challenge, particularly on resource-constrained devices due to the high cost of processing large numbers of visual tokens. Recently, training-free visual token pruning methods have gained popularity as a low-cost solution to this issue. However, existing approaches suffer from two key limitations: semantic saliency-based strategies primarily focus on high cross-attention visual tokens, often neglecting visual diversity, whereas visual diversity-based methods risk inadvertently discarding semantically important tokens, especially under high compression ratios. In this paper, we introduce GreedyPrune, a training-free plug-and-play visual token pruning algorithm designed to jointly optimize semantic saliency and visual diversity. We formalize the token pruning process as a combinatorial optimization problem and demonstrate that greedy algorithms effectively balance computational efficiency with model accuracy. Extensive experiments validate the effectiveness of our approach, showing that GreedyPrune achieves state-of-the-art accuracy across various multimodal tasks and models while significantly reducing end-to-end inference latency.

Dan Peng, Zhihui Fu, Zewen Ye et al.

Sparse attention methods exploit the inherent sparsity in attention to speed up the prefilling phase of long-context inference, mitigating the quadratic complexity of full attention computation. While existing sparse attention methods rely on predefined patterns or inaccurate estimations to approximate attention behavior, they often fail to fully capture the true dynamics of attention, resulting in reduced efficiency and compromised accuracy. Instead, we propose a highly accurate sparse attention mechanism that shares similar yet precise attention patterns across heads, enabling a more realistic capture of the dynamic behavior of attention. Our approach is grounded in two key observations: (1) attention patterns demonstrate strong inter-head similarity, and (2) this similarity remains remarkably consistent across diverse inputs. By strategically sharing computed accurate patterns across attention heads, our method effectively captures actual patterns while requiring full attention computation for only a small subset of heads. Comprehensive evaluations demonstrate that our approach achieves superior or comparable speedup relative to state-of-the-art methods while delivering the best overall accuracy.

Hongyu Li, Xiaogang Huang, Zhihui Fu et al.

Preserving original noise residuals in images are critical to image fraud identification. Since the resizing operation during deep learning will damage the microstructures of image noise residuals, we propose a framework for directly training images of original input scales without resizing. Our arbitrary-sized image training method mainly depends on the pseudo-batch gradient descent (PBGD), which bridges the gap between the input batch and the update batch to assure that model updates can normally run for arbitrary-sized images. In addition, a 3-phase alternate training strategy is designed to learn optimal residual kernels for image fraud identification. With the learnt residual kernels and PBGD, the proposed framework achieved the state-of-the-art results in image fraud identification, especially for images with small tampered regions or unseen images with different tampering distributions.

Tianqi Han, Zhihui Fu, Hongyu Li

Recognizing apparel attributes has recently drawn great interest in the computer vision community. Methods based on various deep neural networks have been proposed for image classification, which could be applied to apparel attributes recognition. An interesting problem raised is how to ensemble these methods to further improve the accuracy. In this paper, we propose a two-layer mixture framework for ensemble different networks. In the first layer of this framework, two types of ensemble learning methods, bagging and boosting, are separately applied. Different from traditional methods, our bagging process makes use of the whole training set, not random subsets, to train each model in the ensemble, where several differentiated deep networks are used to promote model variance. To avoid the bias of small-scale samples, the second layer only adopts bagging to mix the results obtained with bagging and boosting in the first layer. Experimental results demonstrate that the proposed mixture framework outperforms any individual network model or either independent ensemble method in apparel attributes classification.