Weiye Song

Muhao Xu, Xueying Zhou, Xizhan Gao et al.

Recently, detecting logical anomalies is becoming a more challenging task compared to detecting structural ones. Existing encoder decoder based methods typically compress inputs into low-dimensional bottlenecks on the assumption that the compression process can effectively suppress the transmission of logical anomalies to the decoder. However, logical anomalies present a particular difficulty because, while their local features often resemble normal semantics, their global semantics deviate significantly from normal patterns. Thanks to the generalisation capabilities inherent in neural networks, these abnormal semantic features can propagate through low-dimensional bottlenecks. This ultimately allows the decoder to reconstruct anomalous images with misleading fidelity. To tackle the above challenge, we propose a novel normality prior guided multi-semantic fusion network for unsupervised anomaly detection. Instead of feeding the compressed bottlenecks to the decoder directly, we introduce the multi-semantic features of normal samples into the reconstruction process. To this end, we first extract abstract global semantics of normal cases by a pre-trained vision-language network, then the learnable semantic codebooks are constructed to store representative feature vectors of normal samples by vector quantisation. Finally, the above multi-semantic features are fused and employed as input to the decoder to guide the reconstruction of anomalies to approximate normality. Extensive experiments are conducted to validate the effectiveness of our proposed method, and it achieves the SOTA performance on the MVTec LOCO AD dataset with improvements of 5.7% in pixel-sPRO and 2.6% in image-AUROC. The source code is available at https://github.com/Xmh-L/NPGMF.

Baochen Fu, Yuntao Du, Cheng Chang et al.

As real-world knowledge continues to evolve, the parametric knowledge acquired by multimodal models during pretraining becomes increasingly difficult to remain consistent with real-world knowledge. Existing research on multimodal knowledge updating focuses only on learning previously unknown knowledge, while overlooking the need to update knowledge that the model has already mastered but that later changes; moreover, evaluation is limited to the same modality, lacking a systematic analysis of cross-modal consistency. To address these issues, this paper proposes MMKU-Bench, a comprehensive evaluation benchmark for multimodal knowledge updating, which contains over 25k knowledge instances and more than 49k images, covering two scenarios, updated knowledge and unknown knowledge, thereby enabling comparative analysis of learning across different knowledge types. On this benchmark, we evaluate a variety of representative approaches, including supervised fine-tuning (SFT), reinforcement learning from human feedback (RLHF), and knowledge editing (KE). Experimental results show that SFT and RLHF are prone to catastrophic forgetting, while KE better preserve general capabilities but exhibit clear limitations in continual updating. Overall, MMKU-Bench provides a reliable and comprehensive evaluation benchmark for multimodal knowledge updating, advancing progress in this field.

Xiang He, Weiye Song, Yiming Wang et al.

Automatic retinal layer segmentation with medical images, such as optical coherence tomography (OCT) images, serves as an important tool for diagnosing ophthalmic diseases. However, it is challenging to achieve accurate segmentation due to low contrast and blood flow noises presented in the images. In addition, the algorithm should be light-weight to be deployed for practical clinical applications. Therefore, it is desired to design a light-weight network with high performance for retinal layer segmentation. In this paper, we propose LightReSeg for retinal layer segmentation which can be applied to OCT images. Specifically, our approach follows an encoder-decoder structure, where the encoder part employs multi-scale feature extraction and a Transformer block for fully exploiting the semantic information of feature maps at all scales and making the features have better global reasoning capabilities, while the decoder part, we design a multi-scale asymmetric attention (MAA) module for preserving the semantic information at each encoder scale. The experiments show that our approach achieves a better segmentation performance compared to the current state-of-the-art method TransUnet with 105.7M parameters on both our collected dataset and two other public datasets, with only 3.3M parameters.