As a pioneering work exploring transformer architecture for 3D point cloud understanding, Point Transformer achieves impressive results on multiple highly competitive benchmarks. In this work, we analyze the limitations of the Point Transformer and propose our powerful and efficient Point Transformer V2 model with novel designs that overcome the limitations of previous work. In particular, we first propose group vector attention, which is more effective than the previous version of vector attention. Inheriting the advantages of both learnable weight encoding and multi-head attention, we present a highly effective implementation of grouped vector attention with a novel grouped weight encoding layer. We also strengthen the position information for attention by an additional position encoding multiplier. Furthermore, we design novel and lightweight partition-based pooling methods which enable better spatial alignment and more efficient sampling. Extensive experiments show that our model achieves better performance than its predecessor and achieves state-of-the-art on several challenging 3D point cloud understanding benchmarks, including 3D point cloud segmentation on ScanNet v2 and S3DIS and 3D point cloud classification on ModelNet40. Our code will be available at https://github.com/Gofinge/PointTransformerV2.
Transformers flexibly operate over sets of real-valued vectors representing task-specific entities and their attributes, where each vector might encode one word-piece token and its position in a sequence, or some piece of information that carries no position at all. But as set processors, transformers are at a disadvantage in reasoning over more general graph-structured data where nodes represent entities and edges represent relations between entities. To address this shortcoming, we generalize transformer attention to consider and update edge vectors in each transformer layer. We evaluate this relational transformer on a diverse array of graph-structured tasks, including the large and challenging CLRS Algorithmic Reasoning Benchmark. There, it dramatically outperforms state-of-the-art graph neural networks expressly designed to reason over graph-structured data. Our analysis demonstrates that these gains are attributable to relational attention's inherent ability to leverage the greater expressivity of graphs over sets.
Monocular visual odometry consists of the estimation of the position of an agent through images of a single camera, and it is applied in autonomous vehicles, medical robots, and augmented reality. However, monocular systems suffer from the scale ambiguity problem due to the lack of depth information in 2D frames. This paper contributes by showing an application of the dense prediction transformer model for scale estimation in monocular visual odometry systems. Experimental results show that the scale drift problem of monocular systems can be reduced through the accurate estimation of the depth map by this model, achieving competitive state-of-the-art performance on a visual odometry benchmark.
We tackle the problem of sequential brick assembly with LEGO bricks to create combinatorial 3D structures. This problem is challenging since this brick assembly task encompasses the characteristics of combinatorial optimization problems. In particular, the number of assemblable structures increases exponentially as the number of bricks used increases. To solve this problem, we propose a new method to predict the scores of the next brick position by employing a U-shaped sparse 3D convolutional neural network. Along with the 3D convolutional network, a one-initialized brick-sized convolution filter is used to efficiently validate assembly constraints between bricks without training itself. By the nature of this one-initialized convolution filter, we can readily consider several different brick types by benefiting from modern implementation of convolution operations. To generate a novel structure, we devise a sampling strategy to determine the next brick position considering the satisfaction of assembly constraints. Moreover, our method is designed for either budget-free or budget-aware scenario where a budget may confine the number of bricks and their types. We demonstrate that our method successfully generates a variety of brick structures and outperforms existing methods with Bayesian optimization, deep graph generative model, and reinforcement learning.
The design choices in the Transformer attention mechanism, including weak inductive bias and quadratic computational complexity, have limited its application for modeling long sequences. In this paper, we introduce Mega, a simple, theoretically grounded, single-head gated attention mechanism equipped with (exponential) moving average to incorporate inductive bias of position-aware local dependencies into the position-agnostic attention mechanism. We further propose a variant of Mega that offers linear time and space complexity yet yields only minimal quality loss, by efficiently splitting the whole sequence into multiple chunks with fixed length. Extensive experiments on a wide range of sequence modeling benchmarks, including the Long Range Arena, neural machine translation, auto-regressive language modeling, and image and speech classification, show that Mega achieves significant improvements over other sequence models, including variants of Transformers and recent state space models.
Human cognition has compositionality. We understand a scene by decomposing the scene into different concepts (e.g., shape and position of an object) and learning the respective laws of these concepts, which may be either natural (e.g., laws of motion) or man-made (e.g., laws of a game). The automatic parsing of these laws indicates the model's ability to understand the scene, which makes law parsing play a central role in many visual tasks. This paper proposes a deep latent variable model for Compositional LAw Parsing (CLAP), which achieves the human-like compositionality ability through an encoding-decoding architecture to represent concepts in the scene as latent variables. CLAP employs concept-specific latent random functions instantiated with Neural Processes to capture the law of concepts. Our experimental results demonstrate that CLAP outperforms the baseline methods in multiple visual tasks such as intuitive physics, abstract visual reasoning, and scene representation. The law manipulation experiments illustrate CLAP's interpretability by modifying specific latent random functions on samples. For example, CLAP learns the laws of position-changing and appearance constancy from the moving balls in a scene, making it possible to exchange laws between samples or compose existing laws into novel laws.
We present DIY-IPS - Do It Yourself - Indoor Positioning System, an open-source real-time indoor positioning mobile application. DIY-IPS detects users' indoor position by employing dual-band RSSI fingerprinting of available WiFi access points. The app can be used, without additional infrastructural costs, to detect users' indoor positions in real time. We published our app as an open source to save other researchers time recreating it. The app enables researchers/users to (1) collect indoor positioning datasets with a ground truth label, (2) customize the app for higher accuracy or other research purposes (3) test the accuracy of modified methods by live testing with ground truth. We ran preliminary experiments to demonstrate the effectiveness of the app.
Position detection of hydraulic cylinder pistons is crucial for numerous industrial automation applications. A typical traditional method is to excite electromagnetic waves in the cylinder structure and analytically solve the piston position based on the scattering parameters measured by a sensor. The core of this approach is a physical model that outlines the relationship between the measured scattering parameters and the targeted piston position. However, this physical model has shortcomings in accuracy and adaptability, especially in extreme conditions. To address these limitations, we propose machine learning and deep learning-based methods to learn the relationship directly in a data-driven manner. As a result, all deep learning models in this paper consistently outperform the physical one by a large margin. We further deliberate on the choice of models based on domain knowledge and provide in-depth analyses combining model performance with real-world physical characteristics. Specifically, we use Convolutional Neural Network (CNN) to discover local interactions of input among adjacent frequencies, apply Complex-Valued Neural Network (CVNN) to exploit the complex-valued nature of electromagnetic scattering parameters, and introduce a novel technique named Frequency Encoding to add weighted frequency information to the model input. The combination of these techniques results in our best-performing model, a complex-valued CNN with Frequency Encoding, which exhibits substantial improvement in accuracy with an error reduction of 1/12 compared to the traditional physical model.
This paper proposes a new method for RTK post-processing. Different from the traditional forward-backward Kalman filter, in our method, the whole system equation is built on a graphical state space model and solved by factor graph optimization. The position solution provided by the forward Kalman filter is used as the linearization points of the graphical state space model. Constant variables, such as double-difference ambiguity, will exist as constants in the graphical state space model, not as time-series variables. It is shown by experiment results that factor graph optimization with a graphical state space model is more effective than Kalman filter with a traditional discrete-time state space model for RTK post-processing problem.
In this paper, we revisit the use of spectrograms in neural networks, by making the window length a continuous parameter optimizable by gradient descent instead of an empirically tuned integer-valued hyperparameter. The contribution is mostly theoretical at this point, but plugging the modified STFT into any existing neural network is straightforward. We first define a differentiable version of the STFT in the case where local bins centers are fixed and independent of the window length parameter. We then discuss the more difficult case where the window length affects the position and number of bins. We illustrate the benefits of this new tool on an estimation and a classification problems, showing it can be of interest not only to neural networks but to any STFT-based signal processing algorithm.
Existing top-performance 3D object detectors typically rely on the multi-modal fusion strategy. This design is however fundamentally restricted due to overlooking the modality-specific useful information and finally hampering the model performance. To address this limitation, in this work we introduce a novel modality interaction strategy where individual per-modality representations are learned and maintained throughout for enabling their unique characteristics to be exploited during object detection. To realize this proposed strategy, we design a DeepInteraction architecture characterized by a multi-modal representational interaction encoder and a multi-modal predictive interaction decoder. Experiments on the large-scale nuScenes dataset show that our proposed method surpasses all prior arts often by a large margin. Crucially, our method is ranked at the first position at the highly competitive nuScenes object detection leaderboard.
Topology-imbalance is a graph-specific imbalance problem caused by the uneven topology positions of labeled nodes, which significantly damages the performance of GNNs. What topology-imbalance means and how to measure its impact on graph learning remain under-explored. In this paper, we provide a new understanding of topology-imbalance from a global view of the supervision information distribution in terms of under-reaching and over-squashing, which motivates two quantitative metrics as measurements. In light of our analysis, we propose a novel position-aware graph structure learning framework named PASTEL, which directly optimizes the information propagation path and solves the topology-imbalance issue in essence. Our key insight is to enhance the connectivity of nodes within the same class for more supervision information, thereby relieving the under-reaching and over-squashing phenomena. Specifically, we design an anchor-based position encoding mechanism, which better incorporates relative topology position and enhances the intra-class inductive bias by maximizing the label influence. We further propose a class-wise conflict measure as the edge weights, which benefits the separation of different node classes. Extensive experiments demonstrate the superior potential and adaptability of PASTEL in enhancing GNNs' power in different data annotation scenarios.
Liquidity Providers on Automated Market Makers generate millions of USD in transaction fees daily. However, the net value of a Liquidity Position is vulnerable to price changes in the underlying assets in the pool. The dominant measure of loss in a Liquidity Position is Impermanent Loss. Impermanent Loss for Constant Function Market Makers has been widely studied. We propose a new metric to measure Liquidity Position PNL based on price movement from the underlying assets. We show how this new metric more appropriately measures the change in the net value of a Liquidity Position as a function of price movement in the underlying assets. Our second contribution is an algorithm to delta hedge arbitrary Liquidity Positions on both uniform liquidity Automated Market Makers (such as Uniswap v2) and concentrated liquidity Automated Market Makers (such as Uniswap v3) via a combination of derivatives.
A 3D caricature is an exaggerated 3D depiction of a human face. The goal of this paper is to model the variations of 3D caricatures in a compact parameter space so that we can provide a useful data-driven toolkit for handling 3D caricature deformations. To achieve the goal, we propose an MLP-based framework for building a deformable surface model, which takes a latent code and produces a 3D surface. In the framework, a SIREN MLP models a function that takes a 3D position on a fixed template surface and returns a 3D displacement vector for the input position. We create variations of 3D surfaces by learning a hypernetwork that takes a latent code and produces the parameters of the MLP. Once learned, our deformable model provides a nice editing space for 3D caricatures, supporting label-based semantic editing and point-handle-based deformation, both of which produce highly exaggerated and natural 3D caricature shapes. We also demonstrate other applications of our deformable model, such as automatic 3D caricature creation.
Due to the complex layouts of documents, it is challenging to extract information for documents. Most previous studies develop multimodal pre-trained models in a self-supervised way. In this paper, we focus on the embedding learning of word blocks containing text and layout information, and propose UTel, a language model with Unified TExt and Layout pre-training. Specifically, we propose two pre-training tasks: Surrounding Word Prediction (SWP) for the layout learning, and Contrastive learning of Word Embeddings (CWE) for identifying different word blocks. Moreover, we replace the commonly used 1D position embedding with a 1D clipped relative position embedding. In this way, the joint training of Masked Layout-Language Modeling (MLLM) and two newly proposed tasks enables the interaction between semantic and spatial features in a unified way. Additionally, the proposed UTel can process arbitrary-length sequences by removing the 1D position embedding, while maintaining competitive performance. Extensive experimental results show UTel learns better joint representations and achieves superior performance than previous methods on various downstream tasks, though requiring no image modality. Code is available at \url{https://github.com/taosong2019/UTel}.
In the RoboCup Small Size League (SSL), teams are encouraged to propose solutions for executing basic soccer tasks inside the SSL field using only embedded sensing information. Thus, this work proposes an embedded monocular vision approach for detecting objects and estimating relative positions inside the soccer field. Prior knowledge from the environment is exploited by assuming objects lay on the ground, and the onboard camera has its position fixed on the robot. We implemented the proposed method on an NVIDIA Jetson Nano and employed SSD MobileNet v2 for 2D Object Detection with TensorRT optimization, detecting balls, robots, and goals with distances up to 3.5 meters. Ball localization evaluation shows that the proposed solution overcomes the currently used SSL vision system for positions closer than 1 meter to the onboard camera with a Root Mean Square Error of 14.37 millimeters. In addition, the proposed method achieves real-time performance with an average processing speed of 30 frames per second.
We propose united implicit functions (UNIF), a part-based method for clothed human reconstruction and animation with raw scans and skeletons as the input. Previous part-based methods for human reconstruction rely on ground-truth part labels from SMPL and thus are limited to minimal-clothed humans. In contrast, our method learns to separate parts from body motions instead of part supervision, thus can be extended to clothed humans and other articulated objects. Our Partition-from-Motion is achieved by a bone-centered initialization, a bone limit loss, and a section normal loss that ensure stable part division even when the training poses are limited. We also present a minimal perimeter loss for SDF to suppress extra surfaces and part overlapping. Another core of our method is an adjacent part seaming algorithm that produces non-rigid deformations to maintain the connection between parts which significantly relieves the part-based artifacts. Under this algorithm, we further propose "Competing Parts", a method that defines blending weights by the relative position of a point to bones instead of the absolute position, avoiding the generalization problem of neural implicit functions with inverse LBS (linear blend skinning). We demonstrate the effectiveness of our method by clothed human body reconstruction and animation on the CAPE and the ClothSeq datasets.
Pairwise debiasing is one of the most effective strategies in reducing position bias in learning-to-rank (LTR) models. However, limiting the scope of this strategy, are the underlying assumptions required by many pairwise debiasing approaches. In this paper, we develop an approach based on a minimalistic set of assumptions that can be applied to a much broader range of user browsing patterns and arbitrary presentation layouts. We implement the approach as a simplified version of the Unbiased LambdaMART and demonstrate that it retains the underlying unbiasedness property in a wider variety of settings than the original algorithm. Finally, using simulations with "golden" relevance labels, we will show that the simplified version compares favourably with the original Unbiased LambdaMART when the examination of different positions in a ranked list is not assumed to be independent.
We leverage pre-trained language models to solve the task of complex NER for two low-resource languages: Chinese and Spanish. We use the technique of Whole Word Masking(WWM) to boost the performance of masked language modeling objective on large and unsupervised corpora. We experiment with multiple neural network architectures, incorporating CRF, BiLSTMs, and Linear Classifiers on top of a fine-tuned BERT layer. All our models outperform the baseline by a significant margin and our best performing model obtains a competitive position on the evaluation leaderboard for the blind test set.
Indoor Positioning based on Machine Learning has drawn increasing attention both in the academy and the industry as meaningful information from the reference data can be extracted. Many researchers are using supervised, semi-supervised, and unsupervised Machine Learning models to reduce the positioning error and offer reliable solutions to the end-users. In this article, we propose a new architecture by combining Convolutional Neural Network (CNN), Long short-term memory (LSTM) and Generative Adversarial Network (GAN) in order to increase the training data and thus improve the position accuracy. The proposed combination of supervised and unsupervised models was tested in 17 public datasets, providing an extensive analysis of its performance. As a result, the positioning error has been reduced in more than 70% of them.