A recent study has shown a phenomenon called neural collapse in that the within-class means of features and the classifier weight vectors converge to the vertices of a simplex equiangular tight frame at the terminal phase of training for classification. In this paper, we explore the corresponding structures of the last-layer feature centers and classifiers in semantic segmentation. Based on our empirical and theoretical analysis, we point out that semantic segmentation naturally brings contextual correlation and imbalanced distribution among classes, which breaks the equiangular and maximally separated structure of neural collapse for both feature centers and classifiers. However, such a symmetric structure is beneficial to discrimination for the minor classes. To preserve these advantages, we introduce a regularizer on feature centers to encourage the network to learn features closer to the appealing structure in imbalanced semantic segmentation. Experimental results show that our method can bring significant improvements on both 2D and 3D semantic segmentation benchmarks. Moreover, our method ranks 1st and sets a new record (+6.8% mIoU) on the ScanNet200 test leaderboard. Code will be available at https://github.com/dvlab-research/Imbalanced-Learning.

在本文中，我们提出了广义参数对比度学习（GPACO/PACO），该学习在不平衡和平衡数据上都很好地工作。基于理论分析，我们观察到，受监督的对比损失倾向于偏向高频类别，从而增加了学习不平衡的学习难度。我们从优化的角度介绍了一组参数班的可学习中心，以重新平衡。此外，我们在平衡的环境下分析了GPACO/PACO损失。我们的分析表明，GPACO/PACO可以适应地增强同一等级样品的强度，因为将更多的样品与相应的中心一起拉在一起并有益于艰难的示例学习。长尾基准测试的实验表明了长尾识别的新最先进。在完整的Imagenet上，与MAE模型相比，从CNN到接受GPACO损失训练的视觉变压器的模型显示出更好的泛化性能和更强的鲁棒性。此外，GPACO可以应用于语义分割任务，并在4个最受欢迎的基准测试中观察到明显的改进。我们的代码可在https://github.com/dvlab-research/parametric-contrastive-learning上找到。

深度神经网络在严重的类不平衡数据集上的表现不佳。鉴于对比度学习的有希望的表现，我们提出了重新平衡的暹罗对比度采矿（RESCOM）来应对不平衡的识别。基于数学分析和仿真结果，我们声称监督的对比学习在原始批次和暹罗批次水平上都遭受双重失衡问题，这比长尾分类学习更为严重。在本文中，在原始批处理水平上，我们引入了级别平衡的监督对比损失，以分配不同类别的自适应权重。在暹罗批次级别，我们提出了一个级别平衡的队列，该队列维持所有类的键相同。此外，我们注意到，相对于对比度逻辑的不平衡对比损失梯度可以将其分解为阳性和负面因素，易于阳性和易于负面因素将使对比度梯度消失。我们建议有监督的正面和负面对挖掘，以获取信息对的对比度计算并改善表示形式学习。最后，为了大致最大程度地提高两种观点之间的相互信息，我们提出了暹罗平衡的软性软件，并与一阶段训练的对比损失结合。广泛的实验表明，在多个长尾识别基准上，RESCON优于先前的方法。我们的代码和模型可公开可用：https：//github.com/dvlab-research/rescom。

Although deep learning has made remarkable progress in processing various types of data such as images, text and speech, they are known to be susceptible to adversarial perturbations: perturbations specifically designed and added to the input to make the target model produce erroneous output. Most of the existing studies on generating adversarial perturbations attempt to perturb the entire input indiscriminately. In this paper, we propose ExploreADV, a general and flexible adversarial attack system that is capable of modeling regional and imperceptible attacks, allowing users to explore various kinds of adversarial examples as needed. We adapt and combine two existing boundary attack methods, DeepFool and Brendel\&Bethge Attack, and propose a mask-constrained adversarial attack system, which generates minimal adversarial perturbations under the pixel-level constraints, namely ``mask-constraints''. We study different ways of generating such mask-constraints considering the variance and importance of the input features, and show that our adversarial attack system offers users good flexibility to focus on sub-regions of inputs, explore imperceptible perturbations and understand the vulnerability of pixels/regions to adversarial attacks. We demonstrate our system to be effective based on extensive experiments and user study.

Recently the deep learning has shown its advantage in representation learning and clustering for time series data. Despite the considerable progress, the existing deep time series clustering approaches mostly seek to train the deep neural network by some instance reconstruction based or cluster distribution based objective, which, however, lack the ability to exploit the sample-wise (or augmentation-wise) contrastive information or even the higher-level (e.g., cluster-level) contrastiveness for learning discriminative and clustering-friendly representations. In light of this, this paper presents a deep temporal contrastive clustering (DTCC) approach, which for the first time, to our knowledge, incorporates the contrastive learning paradigm into the deep time series clustering research. Specifically, with two parallel views generated from the original time series and their augmentations, we utilize two identical auto-encoders to learn the corresponding representations, and in the meantime perform the cluster distribution learning by incorporating a k-means objective. Further, two levels of contrastive learning are simultaneously enforced to capture the instance-level and cluster-level contrastive information, respectively. With the reconstruction loss of the auto-encoder, the cluster distribution loss, and the two levels of contrastive losses jointly optimized, the network architecture is trained in a self-supervised manner and the clustering result can thereby be obtained. Experiments on a variety of time series datasets demonstrate the superiority of our DTCC approach over the state-of-the-art.

Accurate and smooth global navigation satellite system (GNSS) positioning for pedestrians in urban canyons is still a challenge due to the multipath effects and the non-light-of-sight (NLOS) receptions caused by the reflections from surrounding buildings. The recently developed factor graph optimization (FGO) based GNSS positioning method opened a new window for improving urban GNSS positioning by effectively exploiting the measurement redundancy from the historical information to resist the outlier measurements. Unfortunately, the FGO-based GNSS standalone positioning is still challenged in highly urbanized areas. As an extension of the previous FGO-based GNSS positioning method, this paper exploits the potential of the pedestrian dead reckoning (PDR) model in FGO to improve the GNSS standalone positioning performance in urban canyons. Specifically, the relative motion of the pedestrian is estimated based on the raw acceleration measurements from the onboard smartphone inertial measurement unit (IMU) via the PDR algorithm. Then the raw GNSS pseudorange, Doppler measurements, and relative motion from PDR are integrated using the FGO. Given the context of pedestrian navigation with a small acceleration most of the time, a novel soft motion model is proposed to smooth the states involved in the factor graph model. The effectiveness of the proposed method is verified step-by-step through two datasets collected in dense urban canyons of Hong Kong using smartphone-level GNSS receivers. The comparison between the conventional extended Kalman filter, several existing methods, and FGO-based integration is presented. The results reveal that the existing FGO-based GNSS standalone positioning is highly complementary to the PDR's relative motion estimation. Both improved positioning accuracy and trajectory smoothness are obtained with the help of the proposed method.

In person re-identification (ReID) tasks, many works explore the learning of part features to improve the performance over global image features. Existing methods extract part features in an explicit manner, by either using a hand-designed image division or keypoints obtained with external visual systems. In this work, we propose to learn Discriminative implicit Parts (DiPs) which are decoupled from explicit body parts. Therefore, DiPs can learn to extract any discriminative features that can benefit in distinguishing identities, which is beyond predefined body parts (such as accessories). Moreover, we propose a novel implicit position to give a geometric interpretation for each DiP. The implicit position can also serve as a learning signal to encourage DiPs to be more position-equivariant with the identity in the image. Lastly, a set of attributes and auxiliary losses are introduced to further improve the learning of DiPs. Extensive experiments show that the proposed method achieves state-of-the-art performance on multiple person ReID benchmarks.

We are introducing a multi-scale predictive model for video prediction here, whose design is inspired by the "Predictive Coding" theories and "Coarse to Fine" approach. As a predictive coding model, it is updated by a combination of bottom-up and top-down information flows, which is different from traditional bottom-up training style. Its advantage is to reduce the dependence on input information and improve its ability to predict and generate images. Importantly, we achieve with a multi-scale approach -- higher level neurons generate coarser predictions (lower resolution), while the lower level generate finer predictions (higher resolution). This is different from the traditional predictive coding framework in which higher level predict the activity of neurons in lower level. To improve the predictive ability, we integrate an encoder-decoder network in the LSTM architecture and share the final encoded high-level semantic information between different levels. Additionally, since the output of each network level is an RGB image, a smaller LSTM hidden state can be used to retain and update the only necessary hidden information, avoiding being mapped to an overly discrete and complex space. In this way, we can reduce the difficulty of prediction and the computational overhead. Finally, we further explore the training strategies, to address the instability in adversarial training and mismatch between training and testing in long-term prediction. Code is available at https://github.com/Ling-CF/MSPN.

Crowd counting plays an important role in risk perception and early warning, traffic control and scene statistical analysis. The challenges of crowd counting in highly dense and complex scenes lie in the mutual occlusion of the human body parts, the large variation of the body scales and the complexity of imaging conditions. Deep learning based head detection is a promising method for crowd counting. However the highly concerned object detection networks cannot be well applied to this field for two main reasons. First, most of the existing head detection datasets are only annotated with the center points instead of bounding boxes which is mandatory for the canonical detectors. Second, the sample imbalance has not been overcome yet in highly dense and complex scenes because the existing loss functions calculate the positive loss at a single key point or in the entire target area with the same weight. To address these problems, We propose a novel loss function, called Mask Focal Loss, to unify the loss functions based on heatmap ground truth (GT) and binary feature map GT. Mask Focal Loss redefines the weight of the loss contributions according to the situ value of the heatmap with a Gaussian kernel. For better evaluation and comparison, a new synthetic dataset GTA\_Head is made public, including 35 sequences, 5096 images and 1732043 head labels with bounding boxes. Experimental results show the overwhelming performance and demonstrate that our proposed Mask Focal Loss is applicable to all of the canonical detectors and to various datasets with different GT. This provides a strong basis for surpassing the crowd counting methods based on density estimation.

Despite their impressive performance on diverse tasks, large language models (LMs) still struggle with tasks requiring rich world knowledge, implying the limitations of relying solely on their parameters to encode a wealth of world knowledge. This paper aims to understand LMs' strengths and limitations in memorizing factual knowledge, by conducting large-scale knowledge probing experiments of 10 models and 4 augmentation methods on PopQA, our new open-domain QA dataset with 14k questions. We find that LMs struggle with less popular factual knowledge, and that scaling fails to appreciably improve memorization of factual knowledge in the tail. We then show that retrieval-augmented LMs largely outperform orders of magnitude larger LMs, while unassisted LMs remain competitive in questions about high-popularity entities. Based on those findings, we devise a simple, yet effective, method for powerful and efficient retrieval-augmented LMs, which retrieves non-parametric memories only when necessary. Experimental results show that this significantly improves models' performance while reducing the inference costs.