Image anomaly detection and localization have received widespread interest in the community, and knowledge distillation (KD) has been widely explored. Recently, the reverse distillation (RD) paradigm has successfully mitigated the homogenization of anomaly representations in traditional KD arising from identical or similar teacher-student (T-S) architecture. However, in RD, the lack of an effective means to prevent anomalous patterns in the teacher encoder from being leaked into the student decoder undermines potential modeling discrepancies between the T-S model in anomaly representations. To settle this problem, we proposeREverse distillation with latent Anomaly SuppressiON(REASON) method, preventing the student decoder from receiving anomalous patterns by extra means of anomaly filtering during the inference phase, and thus, the student model can only restore representations of anomaly-free images. Specifically, we construct a Siamese teacher encoder architecture, with one branch extracting features from anomaly-free samples and the other synthesizing anomaly features with spatial noise injection from the latent feature level. Next, we design a latent anomaly suppression module to recover normal features from perturbed anomalous features. In this sense, the follow-up student decoder will receive input without abnormal patterns. Thus, representations of the anomaly-free images can be described well, while those of the anomalous images can be well-differentiated between the T-S model. Furthermore, to enhance the model’s anomaly detection and localization capabilities, we propose multi-granularity KD loss to optimize the student decoder to focus on context and local details. Extensive experiments are performed on three benchmark datasets, i.e., MVTec AD, AeBAD, and OCT2017, and the results show the effectiveness and robustness of our proposed approach, which achieves superiority over the current state-of-the-art methods.

Time series are widely used in many classification and regression tasks. However, numerous time series contain unavoidable missing data, making it challenging to model the temporal dynamics of sequential data. Various data imputation methods have been proposed to infer missing values in time series. Although sequences recorded at fixed time intervals are presented in discrete form, they possess an inherent temporal continuity, which is ignored in most existing approaches. In this paper, we propose an end-to-end Attentive Continuous-Time Generative Adversarial Network (ACGANet) to estimate unobserved values in irregular sequences. ACGANet captures the temporal dynamics by transforming the discrete sequence into the continuous-time flow, thereby modeling the underlying distribution of the real data. Furthermore, ACGANet employs an adversarial learning strategy to alleviate the error introduced by imputed values, with the discriminator distinguishing between real and generated samples. Additionally, ACGANet introduces the log-density of hidden temporal states as an auxiliary loss to further optimize the generator. This allows the model to simultaneously focus on the overall temporal dynamics of the time series and the underlying distribution of the missing data. Extensive experiments on three publicly available real-world datasets demonstrate that ACGANet achieves state-of-the-art performance in imputing incomplete time series. Moreover, both qualitative and quantitative analyses validate the effectiveness of the proposed model.

A GPGPU-based collision detection algorithm is proposed. Firstly, the information of OBB hierarchy tree and triangles of tested objects are mapped into some data textures designed for GPGPU-based calculation, such as triangle vertex textures, bounding box size texture, tree node relationship texture, etc., then these textures are downloaded to GPU to complete the data preparation. Secondly, the whole collision detection is executed on GPU, in which three key contents are fulfilled: reading necessary data from related textures correctly by order coordinate method and index coordinate method, detecting the intersection between triangle and OBB, triangle and triangle through a collision detection index array. Lastly, collision detection results are outputted to a texture by FBO technology and read back to CPU for post-processing. The data transmission between CPU and GPU is only twice, which reduce the time to read data. Testing results show that the detecting speed of the algorithm proposed in this paper is faster than the similar CPU-based algorithm obviously with the increasing complexity of tested objects, while keeps the same precision.