Jonathan Li

Lucas Prado Osco, Qiusheng Wu, Eduardo Lopes de Lemos et al.

Segmentation is an essential step for remote sensing image processing. This study aims to advance the application of the Segment Anything Model (SAM), an innovative image segmentation model by Meta AI, in the field of remote sensing image analysis. SAM is known for its exceptional generalization capabilities and zero-shot learning, making it a promising approach to processing aerial and orbital images from diverse geographical contexts. Our exploration involved testing SAM across multi-scale datasets using various input prompts, such as bounding boxes, individual points, and text descriptors. To enhance the model's performance, we implemented a novel automated technique that combines a text-prompt-derived general example with one-shot training. This adjustment resulted in an improvement in accuracy, underscoring SAM's potential for deployment in remote sensing imagery and reducing the need for manual annotation. Despite the limitations encountered with lower spatial resolution images, SAM exhibits promising adaptability to remote sensing data analysis. We recommend future research to enhance the model's proficiency through integration with supplementary fine-tuning techniques and other networks. Furthermore, we provide the open-source code of our modifications on online repositories, encouraging further and broader adaptations of SAM to the remote sensing domain.

Qiqi Ding, Peng Li, Xuefeng Yan et al.

Snow is one of the toughest adverse weather conditions for object detection (OD). Currently, not only there is a lack of snowy OD datasets to train cutting-edge detectors, but also these detectors have difficulties learning latent information beneficial for detection in snow. To alleviate the two above problems, we first establish a real-world snowy OD dataset, named RSOD. Besides, we develop an unsupervised training strategy with a distinctive activation function, called $Peak \ Act$, to quantitatively evaluate the effect of snow on each object. Peak Act helps grading the images in RSOD into four-difficulty levels. To our knowledge, RSOD is the first quantitatively evaluated and graded snowy OD dataset. Then, we propose a novel Cross Fusion (CF) block to construct a lightweight OD network based on YOLOv5s (call CF-YOLO). CF is a plug-and-play feature aggregation module, which integrates the advantages of Feature Pyramid Network and Path Aggregation Network in a simpler yet more flexible form. Both RSOD and CF lead our CF-YOLO to possess an optimization ability for OD in real-world snow. That is, CF-YOLO can handle unfavorable detection problems of vagueness, distortion and covering of snow. Experiments show that our CF-YOLO achieves better detection results on RSOD, compared to SOTAs. The code and dataset are available at https://github.com/qqding77/CF-YOLO-and-RSOD.

Ali Jamali, Swalpa Kumar Roy, Jonathan Li et al.

In the domain of remote sensing image interpretation, road extraction from high-resolution aerial imagery has already been a hot research topic. Although deep CNNs have presented excellent results for semantic segmentation, the efficiency and capabilities of vision transformers are yet to be fully researched. As such, for accurate road extraction, a deep semantic segmentation neural network that utilizes the abilities of residual learning, HetConvs, UNet, and vision transformers, which is called \texttt{ResUNetFormer}, is proposed in this letter. The developed \texttt{ResUNetFormer} is evaluated on various cutting-edge deep learning-based road extraction techniques on the public Massachusetts road dataset. Statistical and visual results demonstrate the superiority of the \texttt{ResUNetFormer} over the state-of-the-art CNNs and vision transformers for segmentation. The code will be made available publicly at \url{https://github.com/aj1365/ResUNetFormer}.

Binjie Chen, Yunzhou Xia, Yu Zang et al.

The unstructured nature of point clouds demands that local aggregation be adaptive to different local structures. Previous methods meet this by explicitly embedding spatial relations into each aggregation process. Although this coupled approach has been shown effective in generating clear semantics, aggregation can be greatly slowed down due to repeated relation learning and redundant computation to mix directional and point features. In this work, we propose to decouple the explicit modelling of spatial relations from local aggregation. We theoretically prove that basic neighbor pooling operations can too function without loss of clarity in feature fusion, so long as essential spatial information has been encoded in point features. As an instantiation of decoupled local aggregation, we present DeLA, a lightweight point network, where in each learning stage relative spatial encodings are first formed, and only pointwise convolutions plus edge max-pooling are used for local aggregation then. Further, a regularization term is employed to reduce potential ambiguity through the prediction of relative coordinates. Conceptually simple though, experimental results on five classic benchmarks demonstrate that DeLA achieves state-of-the-art performance with reduced or comparable latency. Specifically, DeLA achieves over 90\% overall accuracy on ScanObjectNN and 74\% mIoU on S3DIS Area 5. Our code is available at https://github.com/Matrix-ASC/DeLA .

Ravi Raju, Swayambhoo Jain, Bo Li et al.

Large Language Models (LLMs) have revolutionized the landscape of machine learning, yet current benchmarks often fall short in capturing the diverse behavior of these models in real-world applications. A benchmark's usefulness is determined by its ability to clearly differentiate between models of varying capabilities (separability) and closely align with human preferences. Existing frameworks like Alpaca-Eval 2.0 LC \cite{dubois2024lengthcontrolledalpacaevalsimpleway} and Arena-Hard v0.1 \cite{li2024crowdsourced} are limited by their focus on general-purpose queries and lack of diversity across domains such as law, medicine, and multilingual contexts. In this paper, we address these limitations by introducing a novel data pipeline that curates diverse, domain-specific evaluation sets tailored for LLM-as-a-Judge frameworks. Our approach leverages a combination of manual curation, semi-supervised learning to generate clusters, and stratified sampling to ensure balanced representation across a wide range of domains and languages. The resulting evaluation set, which includes 1573 samples across 14 categories, demonstrates high separability (84\%) across ten top-ranked models, and agreement (84\%) with Chatbot Arena and (0.915) Spearman correlation. The agreement values are 9\% better than Arena Hard and 20\% better than AlpacaEval 2.0 LC, while the Spearman coefficient is 0.7 more than the next best benchmark, showcasing a significant improvement in the usefulness of the benchmark. We further provide an open-source evaluation tool that enables fine-grained analysis of model performance across user-defined categories, offering valuable insights for practitioners. This work contributes to the ongoing effort to enhance the transparency, diversity, and effectiveness of LLM evaluation methodologies.

Dening Lu, Qian Xie, Linlin Xu et al.

Although accurate and fast point cloud classification is a fundamental task in 3D applications, it is difficult to achieve this purpose due to the irregularity and disorder of point clouds that make it challenging to achieve effective and efficient global discriminative feature learning. Lately, 3D Transformers have been adopted to improve point cloud processing. Nevertheless, massive Transformer layers tend to incur huge computational and memory costs. This paper presents a novel hierarchical framework that incorporates convolution with Transformer for point cloud classification, named 3D Convolution-Transformer Network (3DCTN), to combine the strong and efficient local feature learning ability of convolution with the remarkable global context modeling capability of Transformer. Our method has two main modules operating on the downsampling point sets, and each module consists of a multi-scale local feature aggregating (LFA) block and a global feature learning (GFL) block, which are implemented by using Graph Convolution and Transformer respectively. We also conduct a detailed investigation on a series of Transformer variants to explore better performance for our network. Various experiments on ModelNet40 demonstrate that our method achieves state-of-the-art classification performance, in terms of both accuracy and efficiency.

Kyle Gao, Yina Gao, Hongjie He et al.

In March 2020, Neural Radiance Field (NeRF) revolutionized Computer Vision, allowing for implicit, neural network-based scene representation and novel view synthesis. NeRF models have found diverse applications in robotics, urban mapping, autonomous navigation, virtual reality/augmented reality, and more. In August 2023, Gaussian Splatting, a direct competitor to the NeRF-based framework, was proposed, gaining tremendous momentum and overtaking NeRF-based research in terms of interest as the dominant framework for novel view synthesis. We present a comprehensive survey of NeRF papers from the past five years (2020-2025). These include papers from the pre-Gaussian Splatting era, where NeRF dominated the field for novel view synthesis and 3D implicit and hybrid representation neural field learning. We also include works from the post-Gaussian Splatting era where NeRF and implicit/hybrid neural fields found more niche applications. Our survey is organized into architecture and application-based taxonomies in the pre-Gaussian Splatting era, as well as a categorization of active research areas for NeRF, neural field, and implicit/hybrid neural representation methods. We provide an introduction to the theory of NeRF and its training via differentiable volume rendering. We also present a benchmark comparison of the performance and speed of classical NeRF, implicit and hybrid neural representation, and neural field models, and an overview of key datasets.

Kyle Gao, Yina Gao, Hongjie He et al.

In March 2020, Neural Radiance Field (NeRF) revolutionized Computer Vision, allowing for implicit, neural network-based scene representation and novel view synthesis. NeRF models have found diverse applications in robotics, urban mapping, autonomous navigation, virtual reality/augmented reality, and more. In August 2023, Gaussian Splatting, a direct competitor to the NeRF-based framework, was proposed, gaining tremendous momentum and overtaking NeRF-based research in terms of interest as the dominant framework for novel view synthesis. We present a comprehensive survey of NeRF papers from the past five years (2020-2025). These include papers from the pre-Gaussian Splatting era, where NeRF dominated the field for novel view synthesis and 3D implicit and hybrid representation neural field learning. We also include works from the post-Gaussian Splatting era where NeRF and implicit/hybrid neural fields found more niche applications. Our survey is organized into architecture and application-based taxonomies in the pre-Gaussian Splatting era, as well as a categorization of active research areas for NeRF, neural field, and implicit/hybrid neural representation methods. We provide an introduction to the theory of NeRF and its training via differentiable volume rendering. We also present a benchmark comparison of the performance and speed of classical NeRF, implicit and hybrid neural representation, and neural field models, and an overview of key datasets.

Dening Lu, Qian Xie, Mingqiang Wei et al.

Transformers have been at the heart of the Natural Language Processing (NLP) and Computer Vision (CV) revolutions. The significant success in NLP and CV inspired exploring the use of Transformers in point cloud processing. However, how do Transformers cope with the irregularity and unordered nature of point clouds? How suitable are Transformers for different 3D representations (e.g., point- or voxel-based)? How competent are Transformers for various 3D processing tasks? As of now, there is still no systematic survey of the research on these issues. For the first time, we provided a comprehensive overview of increasingly popular Transformers for 3D point cloud analysis. We start by introducing the theory of the Transformer architecture and reviewing its applications in 2D/3D fields. Then, we present three different taxonomies (i.e., implementation-, data representation-, and task-based), which can classify current Transformer-based methods from multiple perspectives. Furthermore, we present the results of an investigation of the variants and improvements of the self-attention mechanism in 3D. To demonstrate the superiority of Transformers in point cloud analysis, we present comprehensive comparisons of various Transformer-based methods for classification, segmentation, and object detection. Finally, we suggest three potential research directions, providing benefit references for the development of 3D Transformers.

Jonathan Li, Rohan Bhambhoria, Xiaodan Zhu

Seeking legal advice is often expensive. Recent advancements in machine learning for solving complex problems can be leveraged to help make legal services more accessible to the public. However, real-life applications encounter significant challenges. State-of-the-art language models are growing increasingly large, making parameter-efficient learning increasingly important. Unfortunately, parameter-efficient methods perform poorly with small amounts of data, which are common in the legal domain (where data labelling costs are high). To address these challenges, we propose parameter-efficient legal domain adaptation, which uses vast unsupervised legal data from public legal forums to perform legal pre-training. This method exceeds or matches the fewshot performance of existing models such as LEGAL-BERT on various legal tasks while tuning only approximately 0.1% of model parameters. Additionally, we show that our method can achieve calibration comparable to existing methods across several tasks. To the best of our knowledge, this work is among the first to explore parameter-efficient methods of tuning language models in the legal domain.

Xun Huang, Ziyu Xu, Hai Wu et al.

LiDAR-based vision systems are integral for 3D object detection, which is crucial for autonomous navigation. However, they suffer from performance degradation in adverse weather conditions due to the quality deterioration of LiDAR point clouds. Fusing LiDAR with the weather-robust 4D radar sensor is expected to solve this problem. However, the fusion of LiDAR and 4D radar is challenging because they differ significantly in terms of data quality and the degree of degradation in adverse weather. To address these issues, we introduce L4DR, a weather-robust 3D object detection method that effectively achieves LiDAR and 4D Radar fusion. Our L4DR includes Multi-Modal Encoding (MME) and Foreground-Aware Denoising (FAD) technique to reconcile sensor gaps, which is the first exploration of the complementarity of early fusion between LiDAR and 4D radar. Additionally, we design an Inter-Modal and Intra-Modal ({IM}2 ) parallel feature extraction backbone coupled with a Multi-Scale Gated Fusion (MSGF) module to counteract the varying degrees of sensor degradation under adverse weather conditions. Experimental evaluation on a VoD dataset with simulated fog proves that L4DR is more adaptable to changing weather conditions. It delivers a significant performance increase under different fog levels, improving the 3D mAP by up to 20.0% over the traditional LiDAR-only approach. Moreover, the results on the K-Radar dataset validate the consistent performance improvement of L4DR in real-world adverse weather conditions.

Jonathan Li, Rohan Bhambhoria, Samuel Dahan et al.

Generative AI models, such as the GPT and Llama series, have significant potential to assist laypeople in answering legal questions. However, little prior work focuses on the data sourcing, inference, and evaluation of these models in the context of laypersons. To this end, we propose a human-centric legal NLP pipeline, covering data sourcing, inference, and evaluation. We introduce and release a dataset, LegalQA, with real and specific legal questions spanning from employment law to criminal law, corresponding answers written by legal experts, and citations for each answer. We develop an automatic evaluation protocol for this dataset, then show that retrieval-augmented generation from only 850 citations in the train set can match or outperform internet-wide retrieval, despite containing 9 orders of magnitude less data. Finally, we propose future directions for open-sourced efforts, which fall behind closed-sourced models.

Quanyun Wu, Kyle Gao, Daniel Long et al.

Embodied AI training and evaluation require object-centric digital twin environments with accurate metric geometry and semantic grounding. Recent transformer-based feedforward reconstruction methods can efficiently predict global point clouds from sparse monocular videos, yet these geometries suffer from inherent scale ambiguity and inconsistent coordinate conventions. This mismatch prevents the reliable fusion of these dimensionless point cloud predictions with locally reconstructed object meshes. We propose a novel scale-aware 3D fusion framework that registers visually grounded object meshes with transformer-predicted global point clouds to construct metrically consistent digital twins. Our method introduces a Vision-Language Model (VLM)-guided geometric anchor mechanism that resolves this fundamental coordinate mismatch by recovering an accurate real-world metric scale. To fuse these networks, we propose a geometry-aware registration pipeline that explicitly enforces physical plausibility through gravity-aligned vertical estimation, Manhattan-world structural constraints, and collision-free local refinement. Experiments on real indoor kitchen environments demonstrate improved cross-network object alignment and geometric consistency for downstream tasks, including multi-primitive fitting and metric measurement. We additionally introduce an open-source indoor digital twin dataset with metrically scaled scenes and semantically grounded and registered object-centric mesh annotations.

Nannan Qin, Pengjie Tao, Haiyan Guan et al.

High-quality digital terrain models derived from airborne laser scanning (ALS) data are essential for a wide range of geospatial analyses, and their generation typically relies on robust ground filtering (GF) to separate point clouds across diverse landscapes into ground and non-ground parts. Although current deep-learning-based GF methods have demonstrated impressive performance, especially in specific challenging terrains, their cross-scene generalization remains limited by two persistent issues: the context-detail dilemma in large-scale processing due to limited computational resources, and the random misclassification of tall objects arising from classification-only optimization. To overcome these limitations, we propose SparseGF, a height-aware sparse segmentation framework enhanced with context compression. It is built upon three key innovations: (1) a convex-mirror-inspired context compression module that condenses expansive contexts into compact representations while preserving central details; (2) a hybrid sparse voxel-point network architecture that effectively interprets compressed representations while mitigating compression-induced geometric distortion; and (3) a height-aware loss function that explicitly enforces topographic elevation priors during training to suppress random misclassification of tall objects. Extensive evaluations on two large-scale ALS benchmark datasets demonstrate that SparseGF delivers robust GF across urban to natural terrains, achieving leading performance in complex urban scenes, competitive results on mixed terrains, and moderate yet non-catastrophic accuracy in densely forested steep areas. This work offers new insights into deep-learning-based GF research and encourages further exploration toward truly cross-scene generalization for large-scale environmental monitoring.

Dening Lu, Kyle Gao, Qian Xie et al.

Although the application of Transformers in 3D point cloud processing has achieved significant progress and success, it is still challenging for existing 3D Transformer methods to efficiently and accurately learn both valuable global features and valuable local features for improved applications. This paper presents a novel point cloud representational learning network, called 3D Dual Self-attention Global Local (GLocal) Transformer Network (3DGTN), for improved feature learning in both classification and segmentation tasks, with the following key contributions. First, a GLocal Feature Learning (GFL) block with the dual self-attention mechanism (i.e., a novel Point-Patch Self-Attention, called PPSA, and a channel-wise self-attention) is designed to efficiently learn the GLocal context information. Second, the GFL block is integrated with a multi-scale Graph Convolution-based Local Feature Aggregation (LFA) block, leading to a Global-Local (GLocal) information extraction module that can efficiently capture critical information. Third, a series of GLocal modules are used to construct a new hierarchical encoder-decoder structure to enable the learning of "GLocal" information in different scales in a hierarchical manner. The proposed framework is evaluated on both classification and segmentation datasets, demonstrating that the proposed method is capable of outperforming many state-of-the-art methods on both classification and segmentation tasks.

Yue Zhou, Jue Chen, Zilun Zhang et al.

Remote sensing (RS) large vision-language models (LVLMs) have shown strong promise across visual grounding (VG) tasks. However, existing RS VG datasets predominantly rely on explicit referring expressions-such as relative position, relative size, and color cues-thereby constraining performance on implicit VG tasks that require scenario-specific domain knowledge. This article introduces DVGBench, a high-quality implicit VG benchmark for drones, covering six major application scenarios: traffic, disaster, security, sport, social activity, and productive activity. Each object provides both explicit and implicit queries. Based on the dataset, we design DroneVG-R1, an LVLM that integrates the novel Implicit-to-Explicit Chain-of-Thought (I2E-CoT) within a reinforcement learning paradigm. This enables the model to take advantage of scene-specific expertise, converting implicit references into explicit ones and thus reducing grounding difficulty. Finally, an evaluation of mainstream models on both explicit and implicit VG tasks reveals substantial limitations in their reasoning capabilities. These findings provide actionable insights for advancing the reasoning capacity of LVLMs for drone-based agents. The code and datasets will be released at https://github.com/zytx121/DVGBench

Lizhang Chen, Jonathan Li, Chen Liang et al.

We introduce training-free looped transformers, in which a lightweight inference-time wrapper loops a contiguous mid-stack block of layers of a frozen checkpoint without additional fine-tuning, continued training, or architectural changes. Unlike prior looped transformer methods that train with the looped structure end-to-end, we retrofit recurrence onto pretrained models at test time. We show that naive block reapplication usually degrades performance, highlighting the importance of the loop application strategy. Motivated by viewing a pre-norm transformer block as a forward Euler step on an ODE, we instead treat looping as a refinement of the same approximation, replacing one large update with smaller damped sub-steps. Across seven dense, sparse MoE, and MLA+MoE model families, our method improves Qwen3-4B-Instruct by +2.64 pp on MMLU-Pro, Qwen3-30B-A3B-Instruct by +1.14 pp on CommonsenseQA, and Moonlight-16B-A3B-Instruct by +1.20 pp on OpenBookQA.

Jonathan Li, Nasim Farahini, Evgenii Iuliugin et al.

The proliferation of 100B+ parameter Large Language Models (LLMs) with 100k+ context length support have resulted in increasing demands for on-chip memory to support large KV caches. Techniques such as StreamingLLM and SnapKV demonstrate how to control KV cache size while maintaining model accuracy. Yet, these techniques are not commonly used within industrial deployments using frameworks like vLLM or SGLang. The reason is twofold: on one hand, the static graphs and continuous batching methodology employed by these frameworks make it difficult to admit modifications to the standard multi-head attention algorithm, while on the other hand, the accuracy implications of such techniques on modern instruction-following and reasoning models are not well understood, obfuscating the need for implementing these techniques. In this paper, we explore these accuracy implications on Llama-3.1-8B-Instruct and DeepSeek-R1, and develop SnapStream, a KV cache compression method that can be deployed at scale. We demonstrate the efficacy of SnapStream in a 16-way tensor-parallel deployment of DeepSeek-671B on SambaNova SN40L accelerators running at 128k context length and up to 1832 tokens per second in a real production setting. SnapStream enables $4\times$ improved on-chip memory usage and introduces minimal accuracy degradation on LongBench-v2, AIME24 and LiveCodeBench. To the best of our knowledge, this is the first implementation of sparse KV attention techniques deployed in a production inference system with static graphs and continuous batching.

Rohan Bhambhoria, Samuel Dahan, Jonathan Li et al.

This study evaluates the performance of general-purpose AI, like ChatGPT, in legal question-answering tasks, highlighting significant risks to legal professionals and clients. It suggests leveraging foundational models enhanced by domain-specific knowledge to overcome these issues. The paper advocates for creating open-source legal AI systems to improve accuracy, transparency, and narrative diversity, addressing general AI's shortcomings in legal contexts.

Lizhang Chen, Jonathan Li, Qi Wang et al.

Mixture-of-Experts (MoE) models rely on balanced expert utilization to fully realize their scalability. However, existing load-balancing methods are largely heuristic and operate on noisy mini-batch assignment statistics, introducing bias relative to population-level objectives. We propose $ϕ$-balancing, a principled framework that directly targets population-level expert balance by minimizing a strictly convex, symmetric, and differentiable potential of the expected routing distribution. Using convex duality, we derive an equivalent min-max formulation and obtain a simple online algorithm via mirror descent, yielding an efficient EMA-based routing adjustment with negligible overhead. Across large-scale pretraining and downstream fine-tuning, $ϕ$-balancing consistently outperforms prior Switch-style and loss-free baselines, demonstrating more stable and effective expert utilization.

Kyle Gao, Liangzhi Li, Hongjie He et al.

Recently released open-source pre-trained foundational image segmentation and object detection models (SAM2+GroundingDINO) allow for geometrically consistent segmentation of objects of interest in multi-view 2D images. Users can use text-based or click-based prompts to segment objects of interest without requiring labeled training datasets. Gaussian Splatting allows for the learning of the 3D representation of a scene's geometry and radiance based on 2D images. Combining Google Earth Studio, SAM2+GroundingDINO, 2D Gaussian Splatting, and our improvements in mask refinement based on morphological operations and contour simplification, we created a pipeline to extract the 3D mesh of any building based on its name, address, or geographic coordinates.

Quanyun Wu, Kyle Gao, Wentao Sun et al.

Accurate quantification of forest coverage and combustible biomass (fuel load) is critical for wildfire risk assessment and ecosystem management. However, traditional methods relying on airborne LiDAR or field surveys are cost-prohibitive and time-intensive, while satellite imagery often lacks the vertical resolution required for canopy volume analysis. This paper proposes a novel, automated pipeline for rapid forest inventory using virtual remote sensing data derived from Google Earth Studio (GES). Our approach first generates low-altitude orbital imagery and camera poses for a target region. For dense 3D reconstruction, we employ Pi-Long, developed within the VGGT-Long framework. This model serves as a scalable extension of the Pi-3 feed-forward Transformer architecture. To address the inherent scale ambiguity in monocular reconstruction, we introduce a metric recovery module that aligns the reconstructed trajectory with GES ground truth poses via Sim(3) Umeyama optimization. The metric-scale point cloud is then orthogonally projected into Bird's-Eye-View (BEV) height and density maps. Finally, we employ a watershed-based segmentation algorithm combined with height variance analysis to classify tree species (conifer vs. broadleaf), calculate Leaf Area Index (LAI), and estimate total fuel load. Experimental results demonstrate that this pipeline offers a scalable, cost-effective alternative to physical scanning, enabling near-real-time estimation of forest biomass with high geometric consistency.

Quanyun Wu, Kyle Gao, Wentao Sun et al.

Accurate measurement of island area and coastline length is crucial for coastal zone monitoring and oceanographic analysis. However, traditional measurement and mapping methods usually rely heavily on orthophotos, expensive airborne depth sensors, or dense ground control points, which face serious limitations of high labor costs, time-consuming efforts, and low operational efficiency in vast and inaccessible open sea environments. To overcome these challenges and break away from the reliance on manual field exploration, this paper proposes a geometrically consistent, real-scale island measurement framework based on pure monocular vision. This project significantly reduces the mapping cost through a fully automated process and achieves high-efficiency measurement without prior GIS data. In our system pipeline, only the geographical coordinates or names of the target area need to be input to obtain a low-altitude surrounding image sequence. After obtaining the point clouds, a lightweight trajectory alignment algorithm (Umeyama) is used to restore the global physical scale, and the scaled model is orthorectified, enabling high-precision area and perimeter extraction directly on the 2D rasterized plane. We have fully verified this pipeline on four islands with different terrain features (covering natural landform islands and islands with complex artificial facilities). The experimental results show that the final measurement error of the system is stable at around 10\%, demonstrating excellent accuracy and robustness. Moreover, this framework has outstanding inference speed, requiring only 70 ms to process a single high-resolution image and generate point clouds, providing a highly practical new paradigm for large-scale marine and coastline

Zhengsen Xu, Lanying Wang, Sibo Cheng et al.

In recent decades, the intensification of wildfire activity in western Canada has resulted in substantial socio-economic and environmental losses. Accurate wildfire risk prediction is hindered by the intrinsic stochasticity of ignition and spread and by nonlinear interactions among fuel conditions, meteorology, climate variability, topography, and human activities, challenging the reliability and interpretability of purely data-driven models. We propose a trustworthy data-driven wildfire risk prediction framework based on long-sequence, multi-scale temporal modeling, which integrates heterogeneous drivers while explicitly quantifying predictive uncertainty and enabling process-level interpretation. Evaluated over western Canada during the record-breaking 2023 and 2024 fire seasons, the proposed model outperforms existing time-series approaches, achieving an F1 score of 0.90 and a PR-AUC of 0.98 with low computational cost. Uncertainty-aware analysis reveals structured spatial and seasonal patterns in predictive confidence, highlighting increased uncertainty associated with ambiguous predictions and spatiotemporal decision boundaries. SHAP-based interpretation provides mechanistic understanding of wildfire controls, showing that temperature-related drivers dominate wildfire risk in both years, while moisture-related constraints play a stronger role in shaping spatial and land-cover-specific contrasts in 2024 compared to the widespread hot and dry conditions of 2023. Data and code are available at https://github.com/SynUW/mmFire.

Nannan Qin, Weikai Tan, Lingfei Ma et al.

Ground filtering has remained a widely studied but incompletely resolved bottleneck for decades in the automatic generation of high-precision digital elevation model, due to the dramatic changes of topography and the complex structures of objects. The recent breakthrough of supervised deep learning algorithms in 3D scene understanding brings new solutions for better solving such problems. However, there are few large-scale and scene-rich public datasets dedicated to ground extraction, which considerably limits the development of effective deep-learning-based ground filtering methods. To this end, we present OpenGF, first Ultra-Large-Scale Ground Filtering dataset covering over 47 $km^2$ of 9 different typical terrain scenes built upon open ALS point clouds of 4 different countries around the world. OpenGF contains more than half a billion finely labeled ground and non-ground points, thousands of times the number of labeled points than the de facto standard ISPRS filtertest dataset. We extensively evaluate the performance of state-of-the-art rule-based algorithms and 3D semantic segmentation networks on our dataset and provide a comprehensive analysis. The results have confirmed the capability of OpenGF to train deep learning models effectively. This dataset is released at https://github.com/Nathan-UW/OpenGF to promote more advancing research for ground filtering and large-scale 3D geographic environment understanding.

Kirill Trapeznikov, Gabriel Mancino-Ball, Jonathan Li et al.

The proliferation of generative video technologies has intensified the need for reliable methods to detect and characterize synthetic media. To address this challenge, we organized the \href{https://safe-video-2025.dsri.org}{SAFE: Synthetic Video Detection Challenge}, co-located with the \textit{Authenticity and Provenance in the Age of Generative AI (APAI) Workshop }at ICCV 2025. The competition invited participants to develop and evaluate algorithms capable of distinguishing real from synthetic videos under fully blind evaluation conditions with over 600 submissions from 12 teams over a 90 day span. Hosted on the Hugging Face platform, the challenge comprised two primary tasks: (1) detection of synthetic video content generated by diverse state-of-the-art models, and (2) detection of synthetic content following common post-processing operations such as resizing, re-compression, motion blur and others. The challenge data consisted of 13 modern high quality synthetic video models with generated content matched to real videos from 21 diverse and challenge sources, all adding up to 20 hours of 6,000 video samples. This paper describes the challenge design, dataset construction, evaluation methodology, and outcomes, offering insights into the generalization and robustness of contemporary synthetic video detection methods. Our findings highlight measurable progress in cross-generator generalization but also persistent vulnerabilities to post-processing artifacts. https://safe-video-2025.dsri.org

Team Cohere, Aakanksha, Arash Ahmadian et al. · mila

In this report we describe the development of Command A, a powerful large language model purpose-built to excel at real-world enterprise use cases. Command A is an agent-optimised and multilingual-capable model, with support for 23 languages of global business, and a novel hybrid architecture balancing efficiency with top of the range performance. It offers best-in-class Retrieval Augmented Generation (RAG) capabilities with grounding and tool use to automate sophisticated business processes. These abilities are achieved through a decentralised training approach, including self-refinement algorithms and model merging techniques. We also include results for Command R7B which shares capability and architectural similarities to Command A. Weights for both models have been released for research purposes. This technical report details our original training pipeline and presents an extensive evaluation of our models across a suite of enterprise-relevant tasks and public benchmarks, demonstrating excellent performance and efficiency.

Lizhang Chen, Jonathan Li, Qiang Liu

The pursuit of faster optimization algorithms remains an active and important research direction in deep learning. Recently, the Muon optimizer [JJB+24] has demonstrated promising empirical performance, but its theoretical foundation remains less understood. In this paper, we bridge this gap and provide a theoretical analysis of Muon by placing it within the Lion-$\mathcal{K}$ family of optimizers [CLLL24]. Specifically, we show that Muon corresponds to Lion-$\mathcal{K}$ when equipped with the nuclear norm, and we leverage the theoretical results of Lion-$\mathcal{K}$ to establish that Muon (with decoupled weight decay) implicitly solves an optimization problem that enforces a constraint on the spectral norm of weight matrices. This perspective not only demystifies the implicit regularization effects of Muon but also leads to natural generalizations through varying the choice of convex map $\mathcal{K}$, allowing for the exploration of a broader class of implicitly regularized and constrained optimization algorithms.

Zoltan Csaki, Bo Li, Jonathan Li et al.

Despite the widespread availability of LLMs, there remains a substantial gap in their capabilities and availability across diverse languages. One approach to address these issues has been to take an existing pre-trained LLM and continue to train it on new languages. While prior works have experimented with language adaptation, many questions around best practices and methodology have not been covered. In this paper, we present a comprehensive investigation into the adaptation of LLMs to new languages. Our study covers the key components in this process, including vocabulary extension, direct preference optimization and the data scarcity problem for human alignment in low-resource languages. We scale these experiments across 9 languages and 2 parameter scales (7B and 70B). We compare our models against Llama 2, Aya-101, XGLM, BLOOM and existing language experts, outperforming all prior published baselines. Additionally, all evaluation code and checkpoints are made public to facilitate future research.

Ling Yue, Jonathan Li, Sixue Xing et al.

The clinical trial process, a critical phase in drug development, is essential for developing new treatments. The primary goal of interventional clinical trials is to evaluate the safety and efficacy of drug-based treatments for specific diseases. However, these trials are often lengthy, labor-intensive, and expensive. The duration of a clinical trial significantly impacts overall costs, making efficient timeline management crucial for controlling budgets and ensuring the economic feasibility of research. To address this issue, We propose TrialDura, a machine learning-based method that estimates the duration of clinical trials using multimodal data, including disease names, drug molecules, trial phases, and eligibility criteria. Then, we encode them into Bio-BERT embeddings specifically tuned for biomedical contexts to provide a deeper and more relevant semantic understanding of clinical trial data. Finally, the model's hierarchical attention mechanism connects all of the embeddings to capture their interactions and predict clinical trial duration. Our proposed model demonstrated superior performance with a mean absolute error (MAE) of 1.04 years and a root mean square error (RMSE) of 1.39 years compared to the other models, indicating more accurate clinical trial duration prediction. Publicly available code can be found at: https://anonymous.4open.science/r/TrialDura-F196.

Kyle Gao, Dening Lu, Hongjie He et al.

3D urban scene reconstruction and modelling is a crucial research area in remote sensing with numerous applications in academia, commerce, industry, and administration. Recent advancements in view synthesis models have facilitated photorealistic 3D reconstruction solely from 2D images. Leveraging Google Earth imagery, we construct a 3D Gaussian Splatting model of the Waterloo region centered on the University of Waterloo and are able to achieve view-synthesis results far exceeding previous 3D view-synthesis results based on neural radiance fields which we demonstrate in our benchmark. Additionally, we retrieved the 3D geometry of the scene using the 3D point cloud extracted from the 3D Gaussian Splatting model which we benchmarked against our Multi- View-Stereo dense reconstruction of the scene, thereby reconstructing both the 3D geometry and photorealistic lighting of the large-scale urban scene through 3D Gaussian Splatting

Guangtao Wang, Shubhangi Upasani, Chen Wu et al.

Efficient long-context inference is critical as large language models (LLMs) adopt context windows of ranging from 128K to 1M tokens. However, the growing key-value (KV) cache and the high computational complexity of attention create significant bottlenecks in memory usage and latency. In this paper, we find that attention in diverse long-context tasks exhibits sparsity, and LLMs implicitly "know" which tokens can be dropped or evicted at the head level after the pre-filling stage. Based on this insight, we propose Self-Attention Guided Eviction~(SAGE-KV), a simple and effective KV eviction cache method for long-context inference. After prefilling, our method performs a one-time top-k selection at both the token and head levels to compress the KV cache, enabling efficient inference with the reduced cache. Evaluations on LongBench and three long-context LLMs (Llama3.1-8B-Instruct-128k, Llama3-8B-Prolong-512k-Instruct, and Qwen2.5-7B-Instruct-128k) show that SAGE-KV maintains accuracy comparable to full attention while significantly improving efficiency. Specifically, SAGE-KV achieves 4x higher memory efficiency with improved accuracy over the static KV cache selection method StreamLLM, and 2x higher memory efficiency with better accuracy than the dynamic KV cache selection method Quest.

Swayambhoo Jain, Ravi Raju, Bo Li et al.

Large Language Models (LLMs) have achieved remarkable advancements, but their monolithic nature presents challenges in terms of scalability, cost, and customization. This paper introduces the Composition of Experts (CoE), a modular compound AI system leveraging multiple expert LLMs. CoE leverages a router to dynamically select the most appropriate expert for a given input, enabling efficient utilization of resources and improved performance. We formulate the general problem of training a CoE and discuss inherent complexities associated with it. We propose a two-step routing approach to address these complexities that first uses a router to classify the input into distinct categories followed by a category-to-expert mapping to obtain desired experts. CoE offers a flexible and cost-effective solution to build compound AI systems. Our empirical evaluation demonstrates the effectiveness of CoE in achieving superior performance with reduced computational overhead. Given that CoE comprises of many expert LLMs it has unique system requirements for cost-effective serving. We present an efficient implementation of CoE leveraging SambaNova SN40L RDUs unique three-tiered memory architecture. CoEs obtained using open weight LLMs Qwen/Qwen2-7B-Instruct, google/gemma-2-9b-it, google/gemma-2-27b-it, meta-llama/Llama-3.1-70B-Instruct and Qwen/Qwen2-72B-Instruct achieve a score of $59.4$ with merely $31$ billion average active parameters on Arena-Hard and a score of $9.06$ with $54$ billion average active parameters on MT-Bench.

Kyle Gao, Dening Lu, Liangzhi Li et al.

The Los Angeles wildfires of January 2025 caused more than 250 billion dollars in damage and lasted for nearly an entire month before containment. Following our previous work, the Digital Twin Building, we modify and leverage the multi-agent large language model framework as well as the cloud-mapping integration to study the air quality during the Los Angeles wildfires. Recent advances in large language models have allowed for out-of-the-box automated large-scale data analysis. We use a multi-agent large language system comprised of an Instructor agent and Worker agents. Upon receiving the users' instructions, the Instructor agent retrieves the data from the cloud platform and produces instruction prompts to the Worker agents. The Worker agents then analyze the data and provide summaries. The summaries are finally input back into the Instructor agent, which then provides the final data analysis. We test this system's capability for data-based policy recommendation by assessing our Instructor-Worker LLM system's health recommendations based on air quality during the Los Angeles wildfires.

Dening Lu, Jun Zhou, Kyle Gao et al.

3D Transformers have achieved great success in point cloud understanding and representation. However, there is still considerable scope for further development in effective and efficient Transformers for large-scale LiDAR point cloud scene segmentation. This paper proposes a novel 3D Transformer framework, named 3D Learnable Supertoken Transformer (3DLST). The key contributions are summarized as follows. Firstly, we introduce the first Dynamic Supertoken Optimization (DSO) block for efficient token clustering and aggregating, where the learnable supertoken definition avoids the time-consuming pre-processing of traditional superpoint generation. Since the learnable supertokens can be dynamically optimized by multi-level deep features during network learning, they are tailored to the semantic homogeneity-aware token clustering. Secondly, an efficient Cross-Attention-guided Upsampling (CAU) block is proposed for token reconstruction from optimized supertokens. Thirdly, the 3DLST is equipped with a novel W-net architecture instead of the common U-net design, which is more suitable for Transformer-based feature learning. The SOTA performance on three challenging LiDAR datasets (airborne MultiSpectral LiDAR (MS-LiDAR) (89.3% of the average F1 score), DALES (80.2% of mIoU), and Toronto-3D dataset (80.4% of mIoU)) demonstrate the superiority of 3DLST and its strong adaptability to various LiDAR point cloud data (airborne MS-LiDAR, aerial LiDAR, and vehicle-mounted LiDAR data). Furthermore, 3DLST also achieves satisfactory results in terms of algorithm efficiency, which is up to 5x faster than previous best-performing methods.

Zhengsen Xu, Jonathan Li, Sibo Cheng et al.

Wildfires pose a significant threat to ecosystems, wildlife, and human communities, leading to habitat destruction, pollutant emissions, and biodiversity loss. Accurate wildfire risk prediction is crucial for mitigating these impacts and safeguarding both environmental and human health. This paper provides a comprehensive review of wildfire risk prediction methodologies, with a particular focus on deep learning approaches combined with remote sensing. We begin by defining wildfire risk and summarizing the geographical distribution of related studies. In terms of data, we analyze key predictive features, including fuel characteristics, meteorological and climatic conditions, socioeconomic factors, topography, and hydrology, while also reviewing publicly available wildfire prediction datasets derived from remote sensing. Additionally, we emphasize the importance of feature collinearity assessment and model interpretability to improve the understanding of prediction outcomes. Regarding methodology, we classify deep learning models into three primary categories: time-series forecasting, image segmentation, and spatiotemporal prediction, and further discuss methods for converting model outputs into risk classifications or probability-adjusted predictions. Finally, we identify the key challenges and limitations of current wildfire-risk prediction models and outline several research opportunities. These include integrating diverse remote sensing data, developing multimodal models, designing more computationally efficient architectures, and incorporating cross-disciplinary methods--such as coupling with numerical weather-prediction models--to enhance the accuracy and robustness of wildfire-risk assessments.

Alnur Ali, Ashutosh Baheti, Jonathan Chang et al.

Developing custom reasoning models via Reinforcement Learning (RL) that can incorporate organization-specific knowledge has great potential to address problems faced by enterprise customers. In many of these problems, the reward function is verifiable, a setting termed RL with Verifiable Rewards (RLVR). We apply RLVR to a popular data science benchmark called BIRD that measures the ability of an AI agent to convert a natural language query for a database to SQL executions. We apply a simple and general-purpose training recipe involving careful prompt and model selection, a warm-up stage using our offline RL approach called TAO, followed by rigorous online RLVR training. With no additional training data beyond the BIRD training set and no use of proprietary models, our very first submission to the BIRD leaderboard reached state-of-the-art accuracy on the private test set: 73.56% without self-consistency and 75.68% with self-consistency. In the latter case, our model also required fewer generations than the second-best approach. While BIRD is only a proxy task, the simplicity of our framework makes it broadly applicable to enterprise domains such as business intelligence, data science, and coding.

Wentao Sun, Hanqing Xu, Quanyun Wu et al.

We introduce PointGauss, a novel point cloud-guided framework for real-time multi-object segmentation in Gaussian Splatting representations. Unlike existing methods that suffer from prolonged initialization and limited multi-view consistency, our approach achieves efficient 3D segmentation by directly parsing Gaussian primitives through a point cloud segmentation-driven pipeline. The key innovation lies in two aspects: (1) a point cloud-based Gaussian primitive decoder that generates 3D instance masks within 1 minute, and (2) a GPU-accelerated 2D mask rendering system that ensures multi-view consistency. Extensive experiments demonstrate significant improvements over previous state-of-the-art methods, achieving performance gains of 1.89 to 31.78% in multi-view mIoU, while maintaining superior computational efficiency. To address the limitations of current benchmarks (single-object focus, inconsistent 3D evaluation, small scale, and partial coverage), we present DesktopObjects-360, a novel comprehensive dataset for 3D segmentation in radiance fields, featuring: (1) complex multi-object scenes, (2) globally consistent 2D annotations, (3) large-scale training data (over 27 thousand 2D masks), (4) full 360° coverage, and (5) 3D evaluation masks.

Kyle Gao, Dening Lu, Liangzhi Li et al.

Urban digital twins are virtual replicas of cities that use multi-source data and data analytics to optimize urban planning, infrastructure management, and decision-making. Towards this, we propose a framework focused on the single-building scale. By connecting to cloud mapping platforms such as Google Map Platforms APIs, by leveraging state-of-the-art multi-agent Large Language Models data analysis using ChatGPT(4o) and Deepseek-V3/R1, and by using our Gaussian Splatting-based mesh extraction pipeline, our Digital Twin Buildings framework can retrieve a building's 3D model, visual descriptions, and achieve cloud-based mapping integration with large language model-based data analytics using a building's address, postal code, or geographic coordinates.

Lizhang Chen, Jonathan Li, Kaizhao Liang et al.

We introduce Cautious Weight Decay (CWD), a one-line, optimizer-agnostic modification that applies weight decay only to parameter coordinates whose signs align with the optimizer update. Unlike standard decoupled decay, which implicitly optimizes a regularized or constrained objective, CWD preserves the original loss and admits a bilevel interpretation: it induces sliding-mode behavior upon reaching the stationary manifold, allowing it to search for locally Pareto-optimal stationary points of the unmodified objective. In practice, CWD is a drop-in change for optimizers such as AdamW, Lion, and Muon, requiring no new hyperparameters or additional tuning. For language model pre-training and ImageNet classification, CWD consistently improves final loss and accuracy at million- to billion-parameter scales.

Pengxin Chen, Zhipeng Luo, Xiaoqi Jiang et al.

Maps have always been an essential component of autonomous driving. With the advancement of autonomous driving technology, both the representation and production process of maps have evolved substantially. The article categorizes the evolution of maps into three stages: High-Definition (HD) maps, Lightweight (Lite) maps, and Implicit maps. For each stage, we provide a comprehensive review of the map production workflow, with highlighting technical challenges involved and summarizing relevant solutions proposed by the academic community. Furthermore, we discuss cutting-edge research advances in map representations and explore how these innovations can be integrated into end-to-end autonomous driving frameworks.

Chengjie Jiang, Yunqi Zhou, Jiafeng Yan et al.

Geospatial pixel reasoning aims to generate segmentation masks in remote sensing imagery directly from natural-language instructions. Most existing approaches follow a paradigm that fine-tunes multimodal large language models under supervision with dense pixel-level masks as ground truth. While effective within the training data distribution, this design suffers from two main drawbacks: (1) the high cost of large-scale dense mask annotation, and (2) the limited generalization capability of supervised fine-tuning in out-of-domain scenarios. To address these issues, we propose GRASP, a structured policy-learning framework that integrates a multimodal large language model with a pretrained segmentation model in a cascaded manner. To enhance generalization, we introduce PRIME, a training paradigm that replaces supervised fine-tuning with reinforcement learning to better align reasoning and grounding behaviors with task objectives. To reduce annotation costs, we design BoP-Rewards, which substitutes dense mask labels with bounding box and positive points. It further verifies outputs through two complementary signals: format, which constrains the reasoning and grounding structure to remain syntactically parsable, and accuracy, which evaluates the quality of predicted boxes and points. For evaluation, we train our method and all baselines on EarthReason and GeoPixInstruct, constructing an in-domain benchmark by merging their test sets. We further release GRASP-1k, a fully out-of-domain benchmark with reasoning-intensive queries, reasoning traces, and fine-grained masks. Experimental results demonstrate state-of-the-art (SOTA) in-domain performance and up to 54\% improvement in out-of-domain scenarios, confirming that reinforcement learning with cost-aware rewards provides a robust and scalable paradigm for geospatial pixel reasoning. All code and datasets will be released publicly.

Wentao Sun, Quanyun Wu, Hanqing Xu et al.

3D Gaussian Splatting (3DGS) has emerged as a powerful paradigm for explicit 3D scene representation, yet achieving efficient and consistent 3D segmentation remains challenging. Current methods suffer from prohibitive computational costs, limited 3D spatial reasoning, and an inability to track multiple objects simultaneously. We present Segment Any Gaussians Online (SAGOnline), a lightweight and zero-shot framework for real-time 3D segmentation in Gaussian scenes that addresses these limitations through two key innovations: (1) a decoupled strategy that integrates video foundation models (e.g., SAM2) for view-consistent 2D mask propagation across synthesized views; and (2) a GPU-accelerated 3D mask generation and Gaussian-level instance labeling algorithm that assigns unique identifiers to 3D primitives, enabling lossless multi-object tracking and segmentation across views. SAGOnline achieves state-of-the-art performance on NVOS (92.7% mIoU) and Spin-NeRF (95.2% mIoU) benchmarks, outperforming Feature3DGS, OmniSeg3D-gs, and SA3D by 15--1500 times in inference speed (27 ms/frame). Qualitative results demonstrate robust multi-object segmentation and tracking in complex scenes. Our contributions include: (i) a lightweight and zero-shot framework for 3D segmentation in Gaussian scenes, (ii) explicit labeling of Gaussian primitives enabling simultaneous segmentation and tracking, and (iii) the effective adaptation of 2D video foundation models to the 3D domain. This work allows real-time rendering and 3D scene understanding, paving the way for practical AR/VR and robotic applications.

Jonathan Li, Zoltan Csaki, Nidhi Hiremath et al.

Modern VLMs have achieved near-saturation accuracy in English document visual question-answering (VQA). However, this task remains challenging in lower resource languages due to a dearth of suitable training and evaluation data. In this paper we present scalable methods for curating such datasets by focusing on Hungarian, approximately the 17th highest resource language on the internet. Specifically, we present HuDocVQA and HuDocVQA-manual, document VQA datasets that modern VLMs significantly underperform on compared to English DocVQA. HuDocVQA-manual is a small manually curated dataset based on Hungarian documents from Common Crawl, while HuDocVQA is a larger synthetically generated VQA data set from the same source. We apply multiple rounds of quality filtering and deduplication to HuDocVQA in order to match human-level quality in this dataset. We also present HuCCPDF, a dataset of 117k pages from Hungarian Common Crawl PDFs along with their transcriptions, which can be used for training a model for Hungarian OCR. To validate the quality of our datasets, we show how finetuning on a mixture of these datasets can improve accuracy on HuDocVQA for Llama 3.2 11B Instruct by +7.2%. Our datasets and code will be released to the public to foster further research in multilingual DocVQA.

Kezheng Xiong, Haoen Xiang, Qingshan Xu et al.

Point cloud registration, a fundamental task in 3D vision, has achieved remarkable success with learning-based methods in outdoor environments. Unsupervised outdoor point cloud registration methods have recently emerged to circumvent the need for costly pose annotations. However, they fail to establish reliable optimization objectives for unsupervised training, either relying on overly strong geometric assumptions, or suffering from poor-quality pseudo-labels due to inadequate integration of low-level geometric and high-level contextual information. We have observed that in the feature space, latent new inlier correspondences tend to cluster around respective positive anchors that summarize features of existing inliers. Motivated by this observation, we propose a novel unsupervised registration method termed INTEGER to incorporate high-level contextual information for reliable pseudo-label mining. Specifically, we propose the Feature-Geometry Coherence Mining module to dynamically adapt the teacher for each mini-batch of data during training and discover reliable pseudo-labels by considering both high-level feature representations and low-level geometric cues. Furthermore, we propose Anchor-Based Contrastive Learning to facilitate contrastive learning with anchors for a robust feature space. Lastly, we introduce a Mixed-Density Student to learn density-invariant features, addressing challenges related to density variation and low overlap in the outdoor scenario. Extensive experiments on KITTI and nuScenes datasets demonstrate that our INTEGER achieves competitive performance in terms of accuracy and generalizability.

Pengfei Zhang, Chang Li, Yongjun Zhang et al.

Remotely sensed imagery interpretation (RSII) faces the three major problems: (1) objective representation of spatial distribution patterns; (2) edge uncertainty problem caused by downsampling encoder and intrinsic edge noises (e.g., mixed pixel and edge occlusion etc.); and (3) false detection problem caused by geometric registration error in change detection. To solve the aforementioned problems, uncertainty-diffusion-model-based high-Frequency TransFormer network (UDHF2-Net) is the first to be proposed, whose superiorities are as follows: (1) a spatially-stationary-and-non-stationary high-frequency connection paradigm (SHCP) is proposed to enhance the interaction of spatially frequency-wise stationary and non-stationary features to yield high-fidelity edge extraction result. Inspired by HRFormer, SHCP proposes high-frequency-wise stream to replace high-resolution-wise stream in HRFormer through the whole encoder-decoder process with parallel frequency-wise high-to-low streams, so it improves the edge extraction accuracy by continuously remaining high-frequency information; (2) a mask-and-geo-knowledge-based uncertainty diffusion module (MUDM), which is a self-supervised learning strategy, is proposed to improve the edge accuracy of extraction and change detection by gradually removing the simulated spectrum noises based on geo-knowledge and the generated diffused spectrum noises; (3) a frequency-wise semi-pseudo-Siamese UDHF2-Net is the first to be proposed to balance accuracy and complexity for change detection. Besides the aforementioned spectrum noises in semantic segmentation, MUDM is also a self-supervised learning strategy to effectively reduce the edge false change detection from the generated imagery with geometric registration error.

Dening Lu, Jun Zhou, Kyle et al.

Recently, LiDAR point cloud processing and analysis have made great progress due to the development of 3D Transformers. However, existing 3D Transformer methods usually are computationally expensive and inefficient due to their huge and redundant attention maps. They also tend to be slow due to requiring time-consuming point cloud sampling and grouping processes. To address these issues, we propose an efficient point TransFormer with Dynamic Token Aggregating (DTA-Former) for point cloud representation and processing. Firstly, we propose an efficient Learnable Token Sparsification (LTS) block, which considers both local and global semantic information for the adaptive selection of key tokens. Secondly, to achieve the feature aggregation for sparsified tokens, we present the first Dynamic Token Aggregating (DTA) block in the 3D Transformer paradigm, providing our model with strong aggregated features while preventing information loss. After that, a dual-attention Transformer-based Global Feature Enhancement (GFE) block is used to improve the representation capability of the model. Equipped with LTS, DTA, and GFE blocks, DTA-Former achieves excellent classification results via hierarchical feature learning. Lastly, a novel Iterative Token Reconstruction (ITR) block is introduced for dense prediction whereby the semantic features of tokens and their semantic relationships are gradually optimized during iterative reconstruction. Based on ITR, we propose a new W-net architecture, which is more suitable for Transformer-based feature learning than the common U-net design.

Dening Lu, Jun Zhou, Kyle Yilin Gao et al.

Point cloud segmentation is one of the most important tasks in computer vision with widespread scientific, industrial, and commercial applications. The research thereof has resulted in many breakthroughs in 3D object and scene understanding. Previous methods typically utilized hierarchical architectures for feature representation. However, the commonly used sampling and grouping methods in hierarchical networks are only based on point-wise three-dimensional coordinates, ignoring local semantic homogeneity of point clusters. Additionally, the prevalent Farthest Point Sampling (FPS) method is often a computational bottleneck. To address these issues, we propose a novel 3D point cloud representation network, called Dynamic Clustering Transformer Network (DCTNet). It has an encoder-decoder architecture, allowing for both local and global feature learning. Specifically, we propose novel semantic feature-based dynamic sampling and clustering methods in the encoder, which enables the model to be aware of local semantic homogeneity for local feature aggregation. Furthermore, in the decoder, we propose an efficient semantic feature-guided upsampling method. Our method was evaluated on an object-based dataset (ShapeNet), an urban navigation dataset (Toronto-3D), and a multispectral LiDAR dataset, verifying the performance of DCTNet across a wide variety of practical engineering applications. The inference speed of DCTNet is 3.8-16.8$\times$ faster than existing State-of-the-Art (SOTA) models on the ShapeNet dataset, while achieving an instance-wise mIoU of $86.6\%$, the current top score. Our method similarly outperforms previous methods on the other datasets, verifying it as the new State-of-the-Art in point cloud segmentation.

Jonathan Li, Will Aitken, Rohan Bhambhoria et al.

Parameter-efficient tuning aims to mitigate the large memory requirements of adapting pretrained language models for downstream tasks. For example, one popular method, prefix-tuning, prepends trainable tokens to sequences while freezing the rest of the model's parameters. Although such models attain comparable performance with fine-tuning when applied to sequences with short to moderate lengths, we show their inferior performance when modelling long sequences. To bridge this gap, we propose prefix-propagation, a simple but effective approach that conditions prefixes on previous hidden states. We empirically demonstrate that prefix-propagation outperforms prefix-tuning across long-document tasks, while using 50% fewer parameters. To further investigate the proposed architecture, we also show its advantage in calibration, and perform additional study on its relationship with kernel attention. To the best of our knowledge, this work is the first to focus on parameter-efficient learning for long-sequence language tasks.

Diogo Nunes Goncalves, Jose Marcato Junior, Pedro Zamboni et al.

Multi-task learning has proven to be effective in improving the performance of correlated tasks. Most of the existing methods use a backbone to extract initial features with independent branches for each task, and the exchange of information between the branches usually occurs through the concatenation or sum of the feature maps of the branches. However, this type of information exchange does not directly consider the local characteristics of the image nor the level of importance or correlation between the tasks. In this paper, we propose a semantic segmentation method, MTLSegFormer, which combines multi-task learning and attention mechanisms. After the backbone feature extraction, two feature maps are learned for each task. The first map is proposed to learn features related to its task, while the second map is obtained by applying learned visual attention to locally re-weigh the feature maps of the other tasks. In this way, weights are assigned to local regions of the image of other tasks that have greater importance for the specific task. Finally, the two maps are combined and used to solve a task. We tested the performance in two challenging problems with correlated tasks and observed a significant improvement in accuracy, mainly in tasks with high dependence on the others.