UniFormer: Unified Transformer for Efficient Spatiotemporal Representation Learning

It is a challenging task to learn rich and multi-scale spatiotemporal semantics from high-dimensional videos, due to large local redundancy and complex global dependency between video frames. The recent advances in this research have been mainly driven by 3D convolutional neural networks and vision transformers. Although 3D convolution can efficiently aggregate local context to suppress local redundancy from a small 3D neighborhood, it lacks the capability to capture global dependency because of the limited receptive field. Alternatively, vision transformers can effectively capture long-range dependency by self-attention mechanism, while having the limitation on reducing local redundancy with blind similarity comparison among all the tokens in each layer. Based on these observations, we propose a novel Unified transFormer (UniFormer) which seamlessly integrates merits of 3D convolution and spatiotemporal self-attention in a concise transformer format, and achieves a preferable balance between computation and accuracy. Different from traditional transformers, our relation aggregator can tackle both spatiotemporal redundancy and dependency, by learning local and global token affinity respectively in shallow and deep layers. We conduct extensive experiments on the popular video benchmarks, e.g., Kinetics-400, Kinetics-600, and Something-Something V1&V2. With only ImageNet-1K pretraining, our UniFormer achieves 82.9%/84.8% top-1 accuracy on Kinetics-400/Kinetics-600, while requiring 10x fewer GFLOPs than other state-of-the-art methods. For Something-Something V1 and V2, our UniFormer achieves new state-of-the-art performances of 60.9% and 71.2% top-1 accuracy respectively. Code is available at https://github.com/Sense-X/UniFormer.

Pedestrian Dead Reckoning System using Quasi-static Magnetic Field Detection

Kalman filter-based Inertial Navigation System (INS) is a reliable and efficient method to estimate the position of a pedestrian indoors. Classical INS-based methodology which is called IEZ (INS-EKF-ZUPT) makes use of an Extended Kalman Filter (EKF), a Zero velocity UPdaTing (ZUPT) to calculate the position and attitude of a person. However, heading error which is a key factor of the whole Pedestrian Dead Reckoning (PDR) system is unobservable for IEZ-based PDR system. To minimize the error, Electronic Com-pass (EC) algorithm becomes a valid method. But magnetic disturbance may have a big negative effect on it. In this paper, the Quasi-static Magnetic field Detection (QMD) method is proposed to detect the pure magnetic field and then selects EC algorithm or Heuristic heading Drift Reduction algorithm (HDR) according to the detection result, which implements the complementation of the two methods. Meanwhile, the QMD, EC, and HDR algorithms are integrated into the IEZ framework to form a new PDR solution which is named Advanced IEZ (AIEZ).

UniFormer: Unifying Convolution and Self-attention for Visual Recognition

It is a challenging task to learn discriminative representation from images and videos, due to large local redundancy and complex global dependency in these visual data. Convolution neural networks (CNNs) and vision transformers (ViTs) have been two dominant frameworks in the past few years. Though CNNs can efficiently decrease local redundancy by convolution within a small neighborhood, the limited receptive field makes it hard to capture global dependency. Alternatively, ViTs can effectively capture long-range dependency via self-attention, while blind similarity comparisons among all the tokens lead to high redundancy. To resolve these problems, we propose a novel Unified transFormer (UniFormer), which can seamlessly integrate the merits of convolution and self-attention in a concise transformer format. Different from the typical transformer blocks, the relation aggregators in our UniFormer block are equipped with local and global token affinity respectively in shallow and deep layers, allowing to tackle both redundancy and dependency for efficient and effective representation learning. Finally, we flexibly stack our UniFormer blocks into a new powerful backbone, and adopt it for various vision tasks from image to video domain, from classification to dense prediction. Without any extra training data, our UniFormer achieves 86.3 top-1 accuracy on ImageNet-1K classification. With only ImageNet-1K pre-training, it can simply achieve state-of-the-art performance in a broad range of downstream tasks, e.g., it obtains 82.9/84.8 top-1 accuracy on Kinetics-400/600, 60.9/71.2 top-1 accuracy on Something-Something V1/V2 video classification tasks, 53.8 box AP and 46.4 mask AP on COCO object detection task, 50.8 mIoU on ADE20K semantic segmentation task, and 77.4 AP on COCO pose estimation task. Code is available at https://github.com/Sense-X/UniFormer.

Uniformer: Unified Transformer for Efficient Spatiotemporal Representation Learning

It is a challenging task to learn rich and multi-scale spatiotemporal semantics from high-dimensional videos, due to large local redundancy and complex global dependency between video frames. The recent advances in this research have been mainly driven by 3D convolutional neural networks and vision transformers. Although 3D convolution can efficiently aggregate local context to suppress local redundancy from a small 3D neighborhood, it lacks the capability to capture global dependency because of the limited receptive field. Alternatively, vision transformers can effectively capture long-range dependency by self-attention mechanism, while having the limitation on reducing local redundancy with blind similarity comparison among all the tokens in each layer. Based on these observations, we propose a novel Unified transFormer (UniFormer) which seamlessly integrates merits of 3D convolution and spatiotemporal self-attention in a concise transformer format, and achieves a preferable balance between computation and accuracy. Different from traditional transformers, our relation aggregator can tackle both spatiotemporal redundancy and dependency, by learning local and global token affinity respectively in shallow and deep layers. We conduct extensive experiments on the popular video benchmarks, e.g., Kinetics-400, Kinetics-600, and Something-Something V1&V2. With only ImageNet-1K pretraining, our UniFormer achieves 82.9%/84.8% top-1 accuracy on Kinetics-400/Kinetics-600, while requiring 10x fewer GFLOPs than other state-of-the-art methods. For Something-Something V1 and V2, our UniFormer achieves new state-of-the-art performances of 60.9% and 71.2% top-1 accuracy respectively. Code is available at https://github.com/Sense-X/UniFormer.

Swift and Sure: Hardness-aware Contrastive Learning for Low-dimensional Knowledge Graph Embeddings

Knowledge graph embedding (KGE) has drawn great attention due to its potential in automatic knowledge graph (KG) completion and knowledge-driven tasks. However, recent KGE models suffer from high training cost and large storage space, thus limiting their practicality in real-world applications. To address this challenge, based on the latest findings in the field of Contrastive Learning, we propose a novel KGE training framework called Hardness-aware Low-dimensional Embedding (HaLE). Instead of the traditional Negative Sampling, we design a new loss function based on query sampling that can balance two important training targets, Alignment and Uniformity. Furthermore, we analyze the hardness-aware ability of recent low-dimensional hyperbolic models and propose a lightweight hardness-aware activation mechanism, which can help the KGE models focus on hard instances and speed up convergence. The experimental results show that in the limited training time, HaLE can effectively improve the performance and training speed of KGE models on five commonly-used datasets. The HaLE-trained models can obtain a high prediction accuracy after training few minutes and are competitive compared to the state-of-the-art models in both low- and high-dimensional conditions.

Ghost-dil-NetVLAD: A Lightweight Neural Network for Visual Place Recognition

Visual place recognition (VPR) is a challenging task with the unbalance between enormous computational cost and high recognition performance. Thanks to the practical feature extraction ability of the lightweight convolution neural networks (CNNs) and the train-ability of the vector of locally aggregated descriptors (VLAD) layer, we propose a lightweight weakly supervised end-to-end neural network consisting of a front-ended perception model called GhostCNN and a learnable VLAD layer as a back-end. GhostCNN is based on Ghost modules that are lightweight CNN-based architectures. They can generate redundant feature maps using linear operations instead of the traditional convolution process, making a good trade-off between computation resources and recognition accuracy. To enhance our proposed lightweight model further, we add dilated convolutions to the Ghost module to get features containing more spatial semantic information, improving accuracy. Finally, rich experiments conducted on a commonly used public benchmark and our private dataset validate that the proposed neural network reduces the FLOPs and parameters of VGG16-NetVLAD by 99.04% and 80.16%, respectively. Besides, both models achieve similar accuracy.

Overcome Anterograde Forgetting with Cycled Memory Networks

Learning from a sequence of tasks for a lifetime is essential for an agent towards artificial general intelligence. This requires the agent to continuously learn and memorize new knowledge without interference. This paper first demonstrates a fundamental issue of lifelong learning using neural networks, named anterograde forgetting, i.e., preserving and transferring memory may inhibit the learning of new knowledge. This is attributed to the fact that the learning capacity of a neural network will be reduced as it keeps memorizing historical knowledge, and the fact that conceptual confusion may occur as it transfers irrelevant old knowledge to the current task. This work proposes a general framework named Cycled Memory Networks (CMN) to address the anterograde forgetting in neural networks for lifelong learning. The CMN consists of two individual memory networks to store short-term and long-term memories to avoid capacity shrinkage. A transfer cell is designed to connect these two memory networks, enabling knowledge transfer from the long-term memory network to the short-term memory network to mitigate the conceptual confusion, and a memory consolidation mechanism is developed to integrate short-term knowledge into the long-term memory network for knowledge accumulation. Experimental results demonstrate that the CMN can effectively address the anterograde forgetting on several task-related, task-conflict, class-incremental and cross-domain benchmarks.

Optimization of RIS Configurations for Multiple-RIS-Aided mmWave Positioning Systems based on CRLB Analysis

Reconfigurable intelligent surface (RIS) is a promising technology for future millimeter-wave (mmWave) communication systems. However, its potential benefits of adopting RIS for high-precision positioning in mmWave systems are still less understood. In this paper, we study a multiple-RIS-aided mmWave positioning system and derive the Cram$\rm{\acute{e}}$r-Rao error bound. Based on the derived bound, we optimize the phase shift of the RISs by the particle swarm optimization (PSO) algorithm. Numerical results have demonstrated the advantages of using multiple RISs in enhancing the positioning accuracy in mmWave systems.

Self-slimmed Vision Transformer

Vision transformers (ViTs) have become the popular structures and outperformed convolutional neural networks (CNNs) on various vision tasks. However, such powerful transformers bring a huge computation burden. And the essential barrier behind this is the exhausting token-to-token comparison. To alleviate this, we delve deeply into the model properties of ViT and observe that ViTs exhibit sparse attention with high token similarity. This intuitively introduces us a feasible structure-agnostic dimension, token number, to reduce the computational cost. Based on this exploration, we propose a generic self-slimmed learning approach for vanilla ViTs, namely SiT. Specifically, we first design a novel Token Slimming Module (TSM), which can boost the inference efficiency of ViTs by dynamic token aggregation. Different from the token hard dropping, our TSM softly integrates redundant tokens into fewer informative ones, which can dynamically zoom visual attention without cutting off discriminative token relations in the images. Furthermore, we introduce a concise Dense Knowledge Distillation (DKD) framework, which densely transfers unorganized token information in a flexible auto-encoder manner. Due to the similar structure between teacher and student, our framework can effectively leverage structure knowledge for better convergence. Finally, we conduct extensive experiments to evaluate our SiT. It demonstrates that our method can speed up ViTs by 1.7x with negligible accuracy drop, and even speed up ViTs by 3.6x while maintaining 97% of their performance. Surprisingly, by simply arming LV-ViT with our SiT, we achieve new state-of-the-art performance on ImageNet, surpassing all the CNNs and ViTs in the recent literature.