Fast Adversarial Training (FastAT) seeks to achieve adversarial robustness at a fraction of the computational cost incurred by standard multi-step methods such as PGD-AT. Although numerous FastAT techniques have been proposed in recent years, fair comparison among them remains elusive. Existing benchmarks and public leaderboards typically permit diverse model architectures, varying training configurations, and external data sources, making it unclear whether reported improvements reflect genuine algorithmic advances or merely more favorable experimental conditions. To address this problem, we introduce the FastAT Benchmark, a controlled evaluation framework built on three core design principles: unified architecture requirements, standardized training settings, and strict prohibition of external or synthetic data. The benchmark implements over twenty representative FastAT methods within a single codebase, enabling direct and reproducible comparison. Each method is assessed through a dual-metric evaluation framework that measures both adversarial robustness (accuracy under PGD, AutoAttack, and CR Attack) and computational cost (GPU training time and peak memory footprint). Comprehensive experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet provide reliable baseline measurements and reveal that well-designed single-step methods can match or surpass PGD-AT robustness at substantially lower cost, while no single method dominates across all evaluation dimensions. The complete benchmark, including source code, configuration files, and experimental results, is publicly available to support transparent and fair evaluation of future FastAT research.

Cloud-hosted LLM inference for autonomous driving adds round-trip delay and depends on stable connectivity, while purely local edge models struggle under occlusion. We present SwarmDrive, a semantic Vehicle-to-Vehicle (V2V) coordination framework in which nearby vehicles run local Small Language Models (SLMs), share compact intent distributions only when uncertainty is high, and fuse them through event-triggered consensus. We evaluate SwarmDrive in a 5-seed executable study built around one occluded intersection case, combining matched operating-point comparisons with robustness sweeps. In that setting, SwarmDrive under its 6G communication setting ("Swarm 6G") raises success from 68.9% to 94.1% over a single local SLM while reducing latency from a 510 ms cloud reference to 151.4 ms. However, an increased number of participating vehicles leads to higher communication overhead and packet loss. SwarmDrive also evaluates the impact of swarm-size, packet-loss, and entropy-threshold sweeps and shows that the cooperative gain holds across ablations and is best balanced near an active swarm size of 4 vehicles and an entropy trigger threshold of 0.65 in the current prototype. These results show that semantic edge cooperation can work under tight latency constraints in the targeted intersection case, but they are not a deployment-grade validation of a real 6G stack.

While Vision-Language Models (VLMs) have advanced highlevel reasoning in autonomous driving, their ability to ground this reasoning in the underlying physics of ego-motion remains poorly understood. We introduce EgoDyn-Bench, a diagnostic benchmark for evaluating the semantic ego-motion understanding of vision-centric foundation models. By mapping continuous vehicle kinematics to discrete motion concepts via a deterministic oracle, we decouple a model's internal physical logic from its visual perception. Our large-scale empirical audit spanning 20 + models, including closed-source MLLMs, open-source VLMs across multiple scales, and specialized VLAs, identifies a significant Perception Bottleneck: while models exhibit logical physical concepts, they consistently fail to accurately align them with visual observations, frequently underperforming classical non-learned geometric baselines. This failure persists across model scales and domain-specific training, indicating a structural deficit in how current architectures couple visual perception with physical reasoning. We demonstrate that providing explicit trajectory encodings substantially restores physical consistency across all evaluated models, revealing a functional disentanglement between vision and language: egomotion logic is derived almost exclusively from the language modality, while visual observations contribute negligible additional signal. This structural finding provides a standardized diagnostic framework and a practical pathway toward physically aligned embodied AI. Keywords: Ego-motion - Physical Reasoning - Foundation Models

Industrial visual inspection systems often suffer from a severe scarcity of labeled defect data, particularly during the early stages of New Product Introduction (NPI). This limitation hinders the deployment of robust supervised detectors precisely when automated quality control is most needed. We present an end-to-end generative framework for high-fidelity, few-shot defect synthesis that enables both in-domain augmentation and cross-domain transfer. Our approach disentangles defect morphology from background appearance by combining masked textual inversion for defect representation learning, noise-blended conditioned generation for surface-aware synthesis, and gradient-aware post-processing for seamless visual integration. We evaluate the framework in two practically relevant settings: few-shot data augmentation, where synthetic samples enrich a small set of real defects, and zero-shot adaptation, where defects learned from a source domain are transferred to a novel target surface without any real target-domain defect examples. Using RF-DETR as the downstream detector, we show that the proposed pipeline substantially narrows the domain gap on a private industrial dataset. In the few-shot setting, synthetic augmentation improves mAP from 78.8% to 83.3%. In the zero-shot setting, synthetic domain adaptation improves mAP from 65.0% to 85.1%. These results demonstrate that high-fidelity defect synthesis can meaningfully accelerate NPI by enabling effective inspection models before sufficient real defect data has been collected.

Recent times have seen an increase in demand of high quality Digital Elevation Models (DEMs) for the lunar surface, because they are highly important for studying the moon and planning future missions. However, there is an evident lack of detailed elevation data on the Moon. To overcome this limitation, this study proposes a novel deep learning method that estimates and generates a surface elevation map directly from monocular images of the surface. The dataset used comprises of the Chandrayaan-2 Terrain Mapping Camera (TMC) images with their corresponding Digital Terrain Models (DTMs). The study proposes LunarDepthNet, which comprises of a UNet architecture to generate DEMS. It incorporates an EfficientNet encoder and custom layers to correctly learn how the light shadows on the surface relate to the actual elevation values. A combined loss function was also utilized to keep the terrain details accurate and smooth. During validation, the model showed a stable loss convergence of 12%. It achieved a mean nRMSE of 0.437 and an MAE of 4.5m in the testing stage. These results prove the model can generate dependable elevation maps from single orbital images, which are quite useful in regions of the moon where stereo-images are not available.

We introduce Dream-Cubed, a large-scale dataset of Minecraft worlds at voxel resolution, and a family of models using cubes as powerful compositional units for efficient generation of interactive 3D environments. Dream-Cubed comprises tens of billions of tokens from a carefully curated mixture of procedural biome terrain and high-quality human-authored maps. We use this dataset to conduct the first large-scale study of 3D diffusion models for voxel generation, analyzing discrete and continuous diffusion formulations, data compositions, and architectural design choices. Our models operate directly in the space of blocks, enabling efficient and semantically grounded generation while supporting interactive user workflows such as inpainting and outpainting from user-authored blocks. To quantitatively evaluate our models, we adapt the FID metric to assess semantic differences between real and generated world renderings, and validate generation quality through a human preference study. We release the full dataset, code, and all our pretrained models, which we hope will provide a foundation for future research in efficient generative modeling for structured, interactive 3D environments.

The expanding ecosystem of pathology foundation models has produced powerful but fragmented tile-level representations, limiting their use in clinical tasks that require unified slide-level reasoning and interpretable linkage to clinically meaningful information. We present ASTRA, a pan-cancer framework that integrates heterogeneous foundation-model representations into a shared slide-level representation space and semantically grounds that space using structured pathology annotation fields, including classification category, cancer type, and anatomic site. ASTRA combines sparse mixture-of-experts contextualization, masked multi-model reconstruction, and contrastive alignment to structured pathology prompts to learn slide representations that support 4-category classification, 3-class solid tumor typing, 16-class cancer typing, and text-guided tumor localization without pixel-level supervision. Developed on a CHTN cohort of 10,359 whole-slide images (WSIs) spanning 16 tumor types, ASTRA consistently improves pan-cancer classification across four pathology foundation-model backbones, achieving up to 97.8% macro-AUC for 4-category classification, 99.7% for 3-class solid tumor typing, and 99.2% for 16-class cancer typing. For tumor localization, ASTRA achieves a mean Dice of 0.897 on an annotated in-domain CHTN subset (n = 380) spanning 16 cancer types and 0.738 on an external TCGA cohort (n = 1,686) spanning four cancer types. These results demonstrate that minimal structured pathology annotation fields derived from slide-level metadata can provide effective semantic supervision for unified slide representation learning, enabling both pan-cancer prediction and weakly supervised tumor localization within a single framework.

Face Image Quality Assessment is crucial for reliable face recognition systems, yet existing Vision Transformer-based approaches rely exclusively on final-layer representations, ignoring quality-relevant information captured at intermediate network depths. This paper presents the first comprehensive investigation of how intermediate representations within ViTs contribute to face quality assessment through early exit mechanisms and score fusion strategies. We systematically analyze all twelve transformer blocks of ViT-FIQA architectures, demonstrating that different depths capture distinct and complementary quality-relevant information, as evidenced by varying attention patterns and performance characteristics across network layers. We propose a score fusion framework that combines quality predictions from multiple transformer blocks without architectural modifications or additional training. Our early exit analysis reveals optimal performance-efficiency trade-offs, enabling significant computational savings while maintaining competitive performance. Through extensive evaluation across eight benchmark datasets using four FR models, we demonstrate that our fusion strategy improves upon single-exit approaches. Our proposed quality fusion approach employs depth-weighted averaging that assigns progressively higher importance to deeper transformer blocks, achieving the best quality assessment performance by effectively leveraging the hierarchical nature of feature learning in ViTs. Our work challenges the conventional wisdom that only deep features matter for face analysis, revealing that intermediate representations contain valuable information for quality assessment. The proposed framework offers practical benefits for real-world biometric systems by enabling adaptive computation based on resource constraints while maintaining competitive quality assessment capabilities.

Face Image Quality Assessment (FIQA) aims to assess the recognition utility of face samples and is essential for reliable face recognition (FR) systems. Existing approaches require computationally expensive procedures such as multiple forward passes, backpropagation, or additional training, and only recent work has focused on the use of Vision Transformers. Recent studies highlighted that these architectures inherently function as saliency learners with attention patterns naturally encoding spatial importance. This work proposes ATTN-FIQA, a novel training-free approach that investigates whether pre-softmax attention scores from pre-trained Vision Transformer-based face recognition models can serve as quality indicators. We hypothesize that attention magnitudes intrinsically encode quality: high-quality images with discriminative facial features enable strong query-key alignments producing focused, high-magnitude attention patterns, while degraded images generate diffuse, low-magnitude patterns. ATTN-FIQA extracts pre-softmax attention matrices from the final transformer block, aggregate multi-head attention information across all patches, and compute image-level quality scores through simple averaging, requiring only a single forward pass through pre-trained models without architectural modifications, backpropagation, or additional training. Through comprehensive evaluation across eight benchmark datasets and four FR models, this work demonstrates that attention-based quality scores effectively correlate with face image quality and provide spatial interpretability, revealing which facial regions contribute most to quality determination.

Large language models (LLMs) have demonstrated strong potential in agentic tasks, particularly in slide generation. However, slide generation poses a fundamental challenge: the generation process is text-centric, whereas its quality is governed by visual aesthetics. This modality gap leads current models to frequently produce slides with aesthetically suboptimal layouts. Existing solutions typically rely either on heavy visual reflection, which incurs high inference cost yet yields limited gains; or on fine-tuning with large-scale datasets, which still provides weak and indirect aesthetic supervision. In contrast, the explicit use of aesthetic principles as supervision remains unexplored. In this work, we present AeSlides, a reinforcement learning framework with verifiable rewards for Aesthetic layout supervision in Slide generation. We introduce a suite of meticulously designed verifiable metrics to quantify slide layout quality, capturing key layout issues in an accurate, efficient, and low-cost manner. Leveraging these verifiable metrics, we develop a GRPO-based reinforcement learning method that directly optimizes slide generation models for aesthetically coherent layouts. With only 5K training prompts on GLM-4.7-Flash, AeSlides improves aspect ratio compliance from 36% to 85%, while reducing whitespace by 44%, element collisions by 43%, and visual imbalance by 28%. Human evaluation further shows a substantial improvement in overall quality, increasing scores from 3.31 to 3.56 (+7.6%), outperforming both model-based reward optimization and reflection-based agentic approaches, and even edging out Claude-Sonnet-4.5. These results demonstrate that such a verifiable aesthetic paradigm provides an efficient and scalable approach to aligning slide generation with human aesthetic preferences. Our repository is available at https://github.com/ympan0508/aeslides.

Precise Event Spotting (PES) is essential in fast-paced sports such as tennis, where fine-grained events occur within very short temporal windows. Accurate frame-level localization is challenging because of motion blur, subtle action differences, and limited annotated data. We study two complementary distillation strategies for few-shot PES: Adaptive Weight Distillation (AWD), a prediction-level method that adaptively weights teacher supervision on unlabeled data, and Annealed Multimodal Distillation for Few-Shot Event Detection (AMD-FED), a representation-level framework that transfers robust skeleton knowledge into visual modalities through annealed pseudo-labeling. Both methods use multimodal distillation to improve generalization under limited supervision. We evaluate them on F3Set-Tennis(sub) under few-shot k-clip settings, where they consistently outperform single-modality baselines and prior PES approaches. After observing the stronger performance of representation-level distillation on tennis, we further validate AMD-FED on a second sports dataset, Figure Skating, where it also shows robust performance in the k-clip scenario. These results highlight the effectiveness of multimodal distillation, especially representation-level transfer, for few-shot precise event spotting.

With the accumulation of resources in the era of big data and the rise of pre-trained models in deep learning, optimizing neural networks for various tasks often involves different strategies for fine-tuning pre-trained models versus training from scratch. However, existing optimizers primarily focus on reducing the loss function by updating model parameters, without fully addressing the unique demands of these two major paradigms. In this paper, we propose DualOpt, a novel approach that decouples optimization techniques specifically tailored for these distinct training scenarios. For training from scratch, we introduce real-time layer-wise weight decay, designed to enhance both convergence and generalization by aligning with the characteristics of weight updates and network architecture. For more importantly fine-tuning, we integrate weight rollback with the optimizer, incorporating a rollback term into each weight update step. This ensures consistency in the weight distribution between upstream and downstream models, effectively mitigating knowledge forgetting and improving fine-tuning performance. Additionally, we extend the layer-wise weight decay to dynamically adjust the rollback levels across layers, adapting to the varying demands of different downstream tasks. Extensive experiments across diverse tasks, including image classification, object detection, semantic segmentation, and instance segmentation, demonstrate the broad applicability and state-of-the-art performance of DualOpt. Code is available at https://github.com/qklee-lz/OLOR-AAAI-2024.

SAM-based dense trackers provide strong short-term mask propagation but remain fragile under long occlusion, fast motion, viewpoint change, and distractors. The problem is especially severe for small objects, where a few incorrect memory updates can dominate later predictions. This report presents an occlusion- and reappearance-aware extension of DAM4SAM that improves memory control rather than changing the backbone. The method augments the original SAM3 tracker with four ingredients: a reliability-aware tracking state machine, branch-based recovery, delayed DRM promotion, and a selective policy for native SAM3 memory selection. During stable tracking, the model follows the original single-path propagation process. Once confidence drops, the tracker enters an ambiguous or recovery mode, maintains a small set of candidate branches, and commits memory only after a branch is reconfirmed. For small-object disappearance and reappearance, native memory selection is temporarily bypassed so older anchors remain accessible. In addition, the first conditioning frame is explicitly preserved, and the conditioning-memory budget is moderately enlarged to improve long-gap recovery. The resulting design keeps DAM4SAM efficient in easy cases while improving robustness in sequences dominated by occlusion and reappearance.

This report describes a Ref-VOS pipeline centered on Sa2VA and organized with explicit agent roles. The key idea is that Sa2VA should provide the first dense semantic hypothesis, while an agent loop decides whether that hypothesis should be accepted, revised, or refined. The pipeline starts with a target-presence judgment stage. If the referred object does not exist in the video, the system directly outputs zero masks. Otherwise, Sa2VA receives the video and referring prompt and produces a coarse mask trajectory over the full video. This trajectory is treated as a semantic prior rather than a final answer. A planner agent decomposes the query, temporal partition agents identify informative blocks, scout agents search for anchor frames, and refinement agents convert reliable Sa2VA masks into boxes and points for SAM3 propagation. A critic scores candidate trajectories, a reflection controller repairs weak hypotheses, and a collaboration controller reconciles multiple agent branches. The result is a Ref-VOS system in which Sa2VA is responsible for dense grounded understanding, while the agent layer handles presence verification, temporal search, confidence-aware revision, and final mask refinement.

Autonomous parking remains a critical yet challenging task in intelligent driving systems, particularly within constrained urban environments where maneuvering space is limited and precise control is essential. While recent advances in end-to-end learning have shown great promise, the lack of high-quality, structured datasets tailored for parking scenarios remains a significant bottleneck.To address this gap, we present ParkingScenes, a comprehensive multimodal dataset specifically designed for end-to-end autonomous parking in simulated scenes. Built on the CARLA simulator, ParkingScenes features structured parking trajectories generated by a Hybrid A* planner and a Model Predictive Controller (MPC), providing accurate and reproducible supervision signals. The dataset includes 16 reverse-in and 6 parallel parking scenarios, each executed under two pedestrian conditions (present and absent), resulting in 704 structured episodes and approximately 105000 frames. Each scenario is repeated 16 times to ensure consistent coverage. Each frame contains synchronized data from four RGB cameras, four depth sensors, vehicle motion states, and Bird's-Eye View (BEV) representations, enabling rich multimodal fusion and context-aware learning. To demonstrate the utility of our dataset, we compare models trained on ParkingScenes with those trained on unstructured, manually collected simulation data under identical conditions. Results show significant improvements in performance, underscoring the effectiveness of structured supervision for robust and accurate parking policy learning. By releasing both the dataset and the collection framework, ParkingScenes establishes a scalable and reproducible benchmark for advancing learning-based autonomous parking systems. The dataset and collection framework will be released at: https://github.com/haonan-ai/ParkingScenes

This paper presents webmcu-vision-web, a single-file, zero-install browser application for end-to-end TinyML vision model training and deployment on the Seeed Studio XIAO ESP32-S3 Sense (XIAO ML Kit, $15--40 USD). Acting as a browser-based companion to the on-device Arduino firmware of Paper 1 [1], it provides a private, fully local machine learning pipeline, from firmware flashing through image collection, CNN training, weight export, and live activation visualization, without any software installation beyond a Chromium-based browser. The system targets educators, small businesses, and researchers who need to train task-specific visual classifiers under their exact deployment conditions. Key capabilities include: in-browser firmware flashing via esptool-js; an SD card file browser with image preview and inline editing; config.json live-sync for zero-recompile hyperparameter adjustment; webcam and ESP32 OV2640 camera image capture; TensorFlow.js CNN training completing a three-class run (~30 images per class, 20 epochs) in approximately 1 minute browser-side versus 9 minutes on-device, enabling a complete collect-train-deploy cycle in under 10 minutes; weight export as myWeights.bin and myWeights.h; confusion matrix; and a live Conv2 activation heatmap streamed from the ESP32 during inference. No data leaves the local machine at any stage. A five-run consistency evaluation on the three-class reference problem (0Blank, 1Cup, 2Pen) demonstrates stable convergence with mean accuracy and standard deviation reported; all artefacts are released at the repository link below. The repository is a living template for LLM-assisted adaptation to new hardware and tasks. All source code is MIT-licensed at https://github.com/webmcu-ai/webmcu-vision-web.

Microscopy-based phenotypic profiling is scalable for drug discovery but lacks the mechanistic depth of transcriptomics, which remains costly and scarce. Existing multimodal approaches either use images to support other modalities or naively align representations by sample identity, ignoring cell-type and dose variations in weakly paired data-limiting generalization to unseen interventions. In this paper, we introduce an intervention-aware distillation framework that leverages perturbational transcriptomics to guide image representation learning. A transcriptome-conditioned teacher integrates gene expression and intervention metadata to produce soft distributions over a chemistry-aware codebook organized by drug similarity. The teacher employs a fine-tuned single-cell foundation model to encode cell-type context and disentangle dose effects. An image-only student learns to predict these distributions from microscopy alone, distilling mechanistic knowledge while operating independently at test time. This design emphasizes intervention semantics rather than identity alignment and explicitly handles dose and cell-type mismatches. We provide theoretical guarantees showing that transcriptomic guidance tightens the risk bound for image-based prediction. On Cell Painting and RxRx datasets paired with L1000, our method significantly improves one-shot transfer to unseen interventions and drug-target gene discovery compared to self-supervised and alignment baselines.

Pre-training is a general method that is used in a range of deep learning tasks. By first training a model on one task, and then further training on the downstream task used for final evaluation, the model is forced to learn a more general understanding of the input data. While pre-training has been applied to 3D Human Pose Estimation (HPE) previously, the scope of datasets used is typically very limited to some strong benchmarks, like Human3.6M. Therefore, in this project, we expand the scope of an existing 3D HPE scheme to be compatible with additional 2D and 3D HPE datasets, like Occlusion Person. We perform an extensive study on how aspects of 2D pre-training, such as model size, affect downstream performance, and to what extent pre-training can help the model generalize to different datasets. Experimental results show that 2D pre-training consistently outperforms training on 3D data alone, particularly in terms of computational efficiency. Finally, using MPII and Human3.6M, we are able to obtain an MPJPE score of under 64.5mm.

The digital transformation of industrial manufacturing increasingly relies on the ability of autonomous robots to interact with legacy infrastructure, particularly analog gauges. While Vision-Language Models (VLMs) have demonstrated potential in zero-shot instrument recognition, their deployment in measurement systems remains constrained by an inherent inability to accurately analyze high-frequency temporal events and needle vibrations. This paper evaluates state-of-the-art models, including GPT-5 and Gemini 3, against the strict requirements of metrology and uncertainty quantification. To facilitate this evaluation, we introduce a novel dataset comprising video sequences of various gauge types: circular, linear, and Vernier, under diverse motion speed profiles. Our findings indicate that current VLMs exhibit limited ability in interpreting needle trajectories and scale semantics, failing to provide the traceability and reliability needed for safety-critical monitoring. The results demonstrate that these models have not yet achieved the performance necessary to be classified as trustworthy synthetic instruments under existing IEEE and ISO standards.

Recent generative AI models have achieved remarkable breakthroughs in language and visual understanding. However, although these models can generate realistic visual content, their spatial scale remains confined to bounded environments, preventing them from capturing how geographic environments evolve across thousands of kilometers or from modeling the spatial structure of the large-scale physical world. This limitation poses a critical challenge for ultra-wide-area spatial intelligence in Earth observation and simulation, revealing a deeper gap in generative AI: progress has relied primarily on scaling model parameters and training data, while overlooking spatial scale as a core dimension of intelligence. Here, motivated by this missing dimension, we investigate spatial scale as a new scaling axis in foundation models and present MetaEarth3D, the first generative foundation model capable of spatially consistent generation at the planetary scale. Taking optical Earth observation simulation as a testbed, MetaEarth3D enables the generation of multi-level, unbounded, and diverse 3D scenes spanning large-scale terrains, medium-scale cities, and fine-grained street blocks. Built upon 10 million globally distributed real-world training images, MetaEarth3D demonstrates both strong visual realism and geospatial statistical realism. Beyond generation, MetaEarth3D serves as a generative data engine for diverse virtual environments in ultra-wide spatial intelligence. We argue that this study may help empower next-generation spatial intelligence for Earth observation.

Millimeter-wave (mmWave) radar has shown great potential for contactless, privacy-preserving, and robust human sensing, yet existing mmWave-based human mesh reconstruction (HMR) studies are still limited by the lack of benchmarks for generalization analysis under configuration shifts and fair comparison of different algorithms. To address the limitation, we present DGHMesh, a large-scale dual-radar mmWave dataset and generalization-focused benchmark for HMR. It contains data from 15 subjects performing 8 actions, with 360,000 synchronized frames collected from FMCW radar, SFCW radar, RGB images, and high-precision 3D HMR annotations. In addition, the dataset provides synchronized raw I/Q data from both radar modalities and accurately calibrated radar spatial positions. The benchmark is designed to evaluate HMR methods under diverse measurement configurations, including human position shifts, human orientation shifts, subarray size variations, and cross-subject settings. Based on DGHMesh, we also propose mmPTM, a query-based multi-radar fusion framework that jointly exploits point clouds and imaging tubes for HMR. Extensive experiments are conducted against representative baselines under different settings. The results demonstrate that mmPTM consistently achieves outstanding accuracy and competitive generalization capability across multiple sub-benchmarks, validating the effectiveness of multi-radar fusion and the practical value of the proposed dataset and benchmark for mmWave-based HMR research. DGHMesh and mmPTM are publicly available at https://github.com/SPIresearch/DGHMesh.(The complete benchmark and code will be released after paper publication)

Industrial CAD workflows require robust, generalizable 3D geometric representations supporting accuracy and explainability. We introduce Shape, a self-supervised foundation model converting surface meshes into dense per-token embeddings. Shape combines a structured 3D latent grid, a multi-scale geometry-aware tokenizer (MAGNO) with cross-attention, and a transformer processor using grouped-query attention and RMSNorm. A learned reconstruction prior enables per-region attribution for explainable predictions. Pretraining uses masked-token reconstruction of normalized geometry statistics and multi-resolution contrastive consistency. The 10.9M-parameter backbone is pretrained on 61,052 CAD meshes from Thingi10K, MFCAD, and Fusion360. On a held-out split of 2,983 meshes, Shape achieves reconstruction R2 = 0.729 and 98.1% top-1 retrieval under the Wang-Isola protocol, with near-zero reconstruction train/val gap (contrastive scores use a larger evaluation pool). A 2x2 ablation on loss type and target-space normalization shows per-dimension normalization is critical: without it, performance collapses (R2 < 0.14, top-1 < 88%); with it, both losses succeed (R2 > 0.70, top-1 > 96%). Smooth-L1 offers secondary stability. Code, embeddings, and an interactive demo are released at https://github.com/simd-ai/shape.

Large segmentation foundation models such as the Segment Anything Model (SAM) have reshaped promptable segmentation in natural images, and recent efforts have extended these models to medical images and volumetric settings. However, directly transferring a 3D SAM-style model to lesion segmentation remains challenging due to (i) weak spatial representational capacity for small, irregular targets in intermediate features, and (ii) extreme foreground-background imbalance in 3D volumes.We propose SGP-SAM, a self-gated prompting framework for efficient and effective transfer to 3D lesion segmentation. Our key component, the Self-Gated Prompting Module (SGPM), performs conditional multi-scale spatial enhancement: a lightweight multi-channel gating unit predicts whether the current features require additional multi-scale fusion, and only then activates a Multi-Scale Feature Fusion Block to enrich spatial context. To further address small-lesion learning, we design a Zoom Loss that up-weights lesion-focused supervision by combining Dice and a voxel-balanced focal term.Experiments on MSD Liver Tumor and MSD Brain Tumor (enhancing tumor) show consistent gains over strong transfer baselines based on SAM-Med3D. On MSD Liver Tumor, SGP-SAM improves mDice by 7.3% over fine-tuning.

Multimodal Large Language Models (MLLMs) rely on multimodal pre-training over diverse data sources, where different datasets often induce complementary cross-modal alignment capabilities. Model merging provides a cost-effective mechanism for integrating multiple expert MLLMs with complementary strengths into a unified model. However, existing model merging research mainly focuses on post-finetuning scenarios, leaving the pre-training stage largely unexplored. We argue that the core of MLLM pre-training lies in establishing effective cross-modal alignment, which bridges visual and textual representations into a unified semantic space. Motivated by this insight, we introduce the post-alignment merging task, which aims to integrate cross-modal alignment capabilities learned from heterogeneous multimodal pre-training. This setting introduces two key challenges: cross-domain parameter interference, where parameter updates learned from different data distributions conflict during merging, and layer-wise alignment contribution disparity, where different layers and projectors contribute unevenly to cross-modal alignment. To address them, we propose \textbf{PivotMerge}, a post-alignment merging framework for cross-modal projectors. PivotMerge incorporates two key components: Shared-space Decomposition and Filtering, which disentangles shared alignment patterns from domain-specific variations and suppresses conflicting directions, and Alignment-guided Layer-wise Merging, which assigns layer-specific merging weights based on differing alignment contributions. We construct systematic CC12M-based post-alignment merging scenarios for evaluation. Extensive experiments on multiple multimodal benchmarks show that PivotMerge consistently outperforms existing baselines, demonstrating its effectiveness and generalization ability.

Object level hallucination remains a central reliability challenge for vision language models (VLMs), particularly in binary object existence verification. Existing benchmarks emphasize aggregate accuracy but rarely disentangle whether errors stem from perceptual limitations or from the influence of contextual textual priors, leaving underlying failure mechanisms ambiguous. We introduce DO-Bench, a controlled diagnostic benchmark that isolates these sources through structured multimodal interventions. Rather than evaluating models in unconstrained settings, DO-Bench probes two complementary dimensions: the Prior Override dimension progressively strengthens contextual textual priors while holding visual evidence constant to assess resistance to prior pressure, and the Perception-Limited dimension incrementally enhances visual evidence from full-scene context to localized object crops to measure perceptual grounding strength. This paired design enables attribution of errors to prior suppression, perceptual insufficiency, or their interaction. We further define two diagnostic metrics, PriorRobust and PerceptionAbility, to quantify these behaviors consistently. Evaluations across diverse open- and closed-source VLMs reveal systematic differences in prior sensitivity and perceptual reliability, demonstrating that object hallucination reflects heterogeneous, mechanism dependent failure patterns beyond aggregate accuracy.

Long-sequence video diffusion transformers hit a quadratic self-attention cost that dominates runtime and memory for very long token sequences. Most efficient attention methods use one approximation everywhere, yet video features are spectrally structured: low frequencies carry global layout and coarse motion; high frequencies carry texture and fine detail. We present FreqFormer, a frequency-aware heterogeneous attention framework. Token features are split into spectral bands with different operators: dense global attention on compressed low-frequency content, structured block-sparse attention on mid frequencies, and sliding-window local attention on high frequencies. A lightweight spectral routing network allocates heads across bands using layer statistics and the diffusion timestep, shifting compute toward global structure early in denoising and detail later. Cross-band summary tokens provide cheap residual exchange. FreqFormer is paired with a fused GPU execution plan that co-schedules dense, sparse, and local branches to cut kernel launches and memory traffic. We give a consistent complexity model, an orthonormal-decomposition view of approximation, and simulation-based systems numbers (throughput, arithmetic intensity, memory traffic, duration scaling). In simulations from 64K to 1M tokens, FreqFormer substantially reduces estimated attention FLOPs and KV-related memory traffic versus dense attention while keeping a hardware-friendly pattern, supporting spectrally structured heterogeneous attention as a practical direction for long-video diffusion transformers.

Augmented reality (AR) systems pose unique privacy risks due to their continuous capture of visual data. Existing AR privacy frameworks lack semantic understanding of visual content, limiting their effectiveness in detecting context-dependent privacy risks. We propose PrivAR, which leverages vision language models (VLMs) with chain-of-thought prompting for contextual privacy risk detection in AR environments. PrivAR uses visual scene cues to infer potential sensitive information types, such as identifying password notes in office environments through contextual reasoning. PrivAR detects and obfuscates textual content, preventing exposure of sensitive information while preserving contextual cues necessary for VLM inference. Additionally, we investigate contextually-informed warning interfaces to enhance user privacy awareness. Experiments on a real-world AR dataset show that PrivAR achieves superior accuracy (81.48%) and F1-score (84.62%) compared to baselines, while reducing privacy leakage rate to 17.58%. User studies evaluating contextually-informed warning interfaces provide insights into effective privacy-aware AR design.

Africa's green industrialization imperative demands reliable infrastructure for monitoring air quality. We present a satellite-reanalysis PM2.5 fusion system trained on 2,068,901 records from 404 monitoring locations in 29 African countries (OpenAQ, 2017-2022), combining LightGBM with leakage-resistant spatial cross-validation and conformal prediction to quantify predictions and their geographic applicability limits. Under 5-fold location-grouped spatial cross-validation, LightGBM achieves RMSE = 30.83 +/- 5.07 ug/m3, MAE = 14.54 +/- 1.66 ug/m3, R2 = 0.134 +/- 0.023, and macro F1 = 0.336 +/- 0.018. This R2 is substantially below random-split benchmarks (>0.90) but reflects true geographic generalisation difficulty rather than model failure. Split conformal prediction targeting 90% marginal coverage reveals severe East Africa degradation (actual PICP = 65.3% vs. nominal 90%), consistent with medium-strength covariate shift (humidity KS = 0.2237, sat_pblh KS = 0.2558). We operationalise these findings through regional reliability flags (High/Medium/Low/Unreliable) and a monitor prioritisation score directing infrastructure expansion toward highest-burden unmonitored populations, directly supporting Africa's green industrial transition and SDGs 3.9, 7.1.2, 9, 11.6.2, and 13.

We present AutoCompress, a transformer compression method motivated by an empirical finding: in small transformers, Layer 0 carries disproportionately high task-critical information, with an NTK-based importance score of 3.6 compared to a maximum of 0.054 for all other layers -- a gap of over 60x. Based on this finding, we propose Critical Layer Isolation (CLI), an architecture that protects Layer 0 at full dimensionality, compresses all intermediate layers through a learned bottleneck, and restores the full dimension at the final layer. Applied to GPT-2 Medium (354.8M parameters), CLI-GPT2 achieves 204.5 perplexity on WikiText-103 with only 143.8M parameters -- a 2.47x compression ratio and 59.5% parameter reduction. Crucially, an ablation study demonstrates that a uniform bottleneck baseline of comparable size achieves only 571.8 perplexity under identical training conditions, confirming that the architectural decision to protect Layer 0 -- rather than simply reducing model size -- is the primary driver of performance. Code and checkpoints are publicly available.

Large language model (LLM) deployments increasingly rely on multi-agent architectures in which multiple models either compete through routing mechanisms or collaborate to produce a final answer. In both settings, the learning signal received by each agent is filtered by the system mechanism. Routing produces selection-gated feedback where only the chosen response is evaluated, while collaboration produces shared rewards that obscure the individual contribution of each agent. As a result, standard RLHF objectives designed for a single deployed policy become misspecified. We introduce CoFi-PGMA (Counterfactual Policy Gradients under Filtered Feedback for Multi-Agent LLMs), a unified framework for learning under filtered feedback in multi-agent LLM systems. Our approach derives a counterfactual per-agent training objective based on marginal contribution, which corrects the learning signal under both routing and collaborative mechanisms. For routing systems, the objective corresponds to off-policy corrections for selection-gated feedback, while for collaborative systems it reduces to leave-one-out difference rewards for credit assignment. We further analyze how softmax routing induces risk-sensitive incentives and provide practical training algorithms that integrate counterfactual estimators, multiturn-aware rewards, and policy optimization methods, and demonstrate the approach on a real-world reasoning dataset.