Papers

This paper describes DS@GT ARC's third-place solution to the PlantCLEF 2026 challenge on multi-species plant identification in vegetation quadrat images, where systems must predict every species present in high-resolution (~3000 x 3000 pixel) plot photographs while training only on single-label images of individual plants. The pipeline is built around a fine-tuned DINOv2 ViT-L/14 classifier applied over a multi-scale tile decomposition of each quadrat, with per-tile predictions blended with a FAISS kNN retriever and post-processed by source-aware temporal fusion across repeated plot visits, a habitat-fit demotion that injects geographic and altitude priors from the training data, and a South-Western Europe geographic mask. Habitat-fit demotion and multi-scale aggregation are the largest individual contributors in the ablations. Two complementary training-centric directions, a cross-region transformer with noisy-student distillation on the LUCAS dataset and a label-as-query transformer decoder over synthetic CLS-domain pseudo-quadrats, yielded null results. An inference-time augmentation with instance-aware segmentation crops also did not improve performance. The selected submission reaches a private-leaderboard macro-F1 of 0.43902 (third place; public 0.51096); an unselected configuration of the same pipeline scored above 0.45 on the private set. Code: https://github.com/dsgt-arc/plantclef-2026.

Egocentric Visual Question Answering (VQA) has attracted widespread attention as an important task for enabling Multimodal Large Language Models (MLLMs) to interact with the real world. However, existing MLLMs struggle to perform effective spatial reasoning in complex egocentric scenes due to their limited spatial perception capabilities. To this end, we introduce Ego Scene Augmentation (ESA), an egocentric spatial perception framework, which actively enhances the spatial perception capabilities from the egocentric perspective, powered by the proposed Ego-element Graph. Our core insight is leveraging the Ego-element Graph as an intermediary representation to augment the egocentric spatial perception of MLLMs via visual foundational models. Specifically, we 1) construct the Ego-element Graph, which encapsulates and integrates egocentric spatial features enabled by visual foundational models; 2) enhance the spatial perception capabilities of MLLMs via the Ego-element Graph for ego-perspective scenes. Our proposed ESA framework presents significant performance improvement on the EgoTextVQA benchmark. We achieve an 8.14% gain on the indoor setting and an 8.72% gain on the outdoor setting. Furthermore, our ESA shows the most impressive performance improvement in the shopping subset of the indoor setting. The project code is publicly available.

3D Gaussian Splatting (3DGS) has become the method of choice for reconstructing and real-time rendering of captured scenes. To capture a scene with good visual quality, continuous image sequences are usually combined with out-of-order shots for better scene coverage. Structure from motion can reconstruct such captures, but only after they are all available and often with high computational cost. Incremental reconstruction methods – often derived from SLAM solutions – provide immediate feedback, but cannot handle the out-of-order capture we require. We provide the first immediate feedback solution for such radiance field capture that provides global consistency. We first introduce a method for fast matching in out-of-order sequences, by repurposing visual place recognition models and a covisibility graph, and provide an efficient way to find highly connected keyframes, improving quality even for ordered sequences. We show how these steps – together with GPU optimization and careful Gaussian primitive placement – provide fast local reconstruction, in our challenging radiance field reconstruction case. We then introduce a novel cluster-based method, again using the covisibility graph, to provide efficient loop closure that does not require sequential input. Finally, to handle large scenes in our context, we introduce a progressive hierarchy that allows our method to scale to large environments, without compromising efficiency. Our results show we provide immediate feedback 3DGS reconstruction with good visual quality in several datasets, with up to thousands of input images.

Few-view surface reconstruction recovers the visible surfaces of a scene from a few posed RGB images, providing the 3D models that robots need to explore and interact online. On mobile platforms, the reconstruction must be fast and geometrically accurate while keeping a small memory footprint to ensure safe and efficient operation. 3D Gaussian Splatting (3DGS) offers a high-fidelity scene representation, but building it from a few views is ill-posed, as many distinct surfaces reproduce the same images, making traditional photometric methods prone to "floater" artifacts. End-to-end methods resolve the ambiguity by regressing splats with large, usually Transformer-based, networks that require heavy compute and memory while generalizing poorly to new scenes. We propose G2SR, which exploits a well-posed core of the task: given cross-view 2D splat correspondences, 3D splats follow analytically from multi-view geometry. G2SR employs a lightweight neural frontend to detect and track 2D Gaussian splats on the image plane and an analytic backend to triangulate each into a metric-scale 3D splat. On ScanNet, Replica, and DTU, G2SR matches or exceeds the geometric accuracy of state-of-the-art end-to-end methods while running at 69-89 reconstructions per second within 203 MB of GPU memory (5-107x less) for 2- and 3-view inputs at 384 x 512 resolution, offering a practical path to online Gaussian-based surface reconstruction.

We study Dynamical System Autoencoders (DSAE) for LiDAR point-cloud classification using spatial coordinates and Product Coefficient feature augmentations. The experiments compare separately trained DSAE architectures at encoder depths $K=1,\ldots,5$ and evaluate the resulting hidden representations with Random Forest, kNN, and a majority-class Dummy baseline. The main finding is a hidden-state collapse at $K=5$. For both xyz and xyz plus Product Coefficient inputs, the hidden-state standard deviation falls to the order of $10^{-5}$, while all three classifiers attain the same macro F1 score of $0.224688$. We prove that between-class hidden scatter is bounded by total hidden scatter, which in turn is controlled by the reported hidden-state variance. Thus a nearly constant hidden representation cannot retain substantial class-separating structure. Product Coefficients neither improve pre-collapse macro F1 nor prevent the $K=5$ collapse in the present DSAE setting. These results identify large-depth representation collapse as a concrete failure mode for DSAE LiDAR classification.

Cotton squares are important phenotypic indicators of the early reproductive growth of cotton, and automatic field detection of cotton squares provides an important basis for cotton growth monitoring and precision cultivation management. However, early cotton square detection in complex field environments remains insufficiently explored, as cotton squares are small, frequently occluded, easily blurred, subject to illumination variations, and exhibit low contrast against surrounding cotton leaves. To address these challenges, we propose a task-oriented framework based on YOLO26m, named Cotton-SF YOLO, for cotton square detection under natural field conditions. To improve the perception of small and irregular cotton square boundaries, we introduce Dynamic Snake Convolution into the detector, enabling adaptive extraction of deformable edge features. Furthermore, a frequency-domain feature modulation module is designed by incorporating spectral enhancement into the C2f structure, which recalibrate frequency-domain representations and strengthen discriminative edge and texture cues while reducing interference from complex cotton leaf backgrounds. Trained and evaluated on our newly constructed and annotated field dataset with manually annotated cotton squares, the proposed model achieves mAP$_{50}$, mAP$_{50:95}$, and recall values of 0.8196, 0.4942, and 0.7939, improving over the baseline YOLO26m by 1.25%, 3.45%, and 2.96%, respectively. Ablation experiments and visualization demonstrate that the best performance is achieved with the complementary effects of structural and frequency cues.

Early detection of dental caries remains challenging due to limitations in traditional diagnostic methods, particularly for proximal lesions in posterior teeth. Deep learning models show promise for automated caries detection but face scalability constraints due to requirements for large volumes of expertly annotated training data. This study presents a dual-stage deep learning framework combining Faster R-CNN for tooth localization with U-Net for pixel-wise caries segmentation in panoramic radiographs. We developed a systematic transformation pipeline to convert large-scale polygon-annotated datasets into high-resolution binary segmentation masks, enabling pixel-wise supervised learning. The framework was trained using both expert-verified datasets and algorithmically processed labels from 3,000 panoramic images. Our approach achieved robust performance with an IoU of 0.9013, Dice coefficient of 0.9482, Recall of 0.9433, and Precision of 0.9774, demonstrating superior accuracy compared to existing methods while significantly reducing false-positive rates. The dual-stage framework effectively addresses data annotation bottlenecks in dental AI applications and demonstrates potential for scalable, automated caries detection systems that can improve diagnostic consistency and support clinical decision-making.

Reducing radiation dose in computed tomography significantly degrades image quality and poses challenges for accurate and clinically reliable reconstruction. While recent approaches have shown promise for low-dose CT, they often struggle to generalize across continuous and previously unseen dose levels, leading to artifacts and loss of anatomical detail. To address these limitations, we propose Dose-Aware Cold Diffusion (DACD), a physics-consistent reconstruction framework that explicitly models radiation dose as a continuous latent factor within a cold diffusion process. The proposed DACD framework integrates image-based dose-aware perception, multi-scale structural prior extraction, and dose-calibrated step allocation to adaptively guide the denoising trajectory. In addition, an iterative forward-backprojection correction is incorporated into the reverse refinement process to enforce projection-domain data consistency. Extensive experiments on three public benchmarks, including Mayo-2020, Mayo-2016, and LoDoPaB-CT, demonstrate that DACD consistently outperforms state-of-the-art diffusion-based and physics-guided methods in both quantitative accuracy and visual fidelity, particularly under ultra-low-dose conditions. The results show that DACD achieves robust generalization across a continuous range of dose levels, including those unseen during training.

Deep learning models for cervical cytology are almost always evaluated as if every prediction must be acted upon, yet a screening system deployed alongside a cytopathologist need not classify every slide: it can defer the cases it is least certain about. Evaluating such a system requires asking not only how often it is correct, but whether its confidence ranks its errors to the bottom. This paper studies selective prediction and uncertainty-aware referral on the Herlev Pap smear dataset under a binary Normal-versus-Abnormal formulation. Two lightweight transformer backbones (Swin-Tiny, TinyViT-5M) are fine-tuned on Herlev from ImageNet-pretrained weights with weighted random sampling, calibrated by post-hoc temperature scaling fit on a held-out calibration subset, and compared against a soft-voting ensemble of both models. Discrimination is reported alongside expected calibration error (ECE) and, as the primary endpoint, the area under the risk-coverage curve (AURC). No statistically significant difference was detected between the two configurations in accuracy or macro-F1, yet the ensemble halves AURC (0.0022 vs. 0.0045, a 51.8% reduction, lower in all five folds) and extends the coverage at which zero errors are made from 18.3% to 72.8% of the pooled test predictions. The same ensemble is nonetheless worse calibrated in absolute terms (ECE 0.0339 vs. 0.0247) and produces more false negatives (14 vs. 10). These results separate two properties that are frequently conflated: the ability to rank predictions by trustworthiness, and the accuracy of the confidence values themselves. Ensembling improves the former while degrading the latter, and the former directly governs the observed risk-coverage tradeoff, whereas the latter governs the interpretation of the reported confidence values.

This paper investigates whether multimodal LLMs can evaluate local brushstroke quality in calligraphy and generate educationally useful natural language feedback. We construct an evaluation framework in which three multimodal LLMs (GPT-4o, Claude Sonnet 4, and Gemini 2.5 Flash) assess before-after image pairs of calligraphic works using a five-point ordinal scale, and compare their outputs against scores assigned by three expert calligraphers. We additionally examine a Retrieval-Augmented Generation (RAG) variant of Claude as a preliminary condition. Results show that all models achieve useful levels of absolute score accuracy (MAE), with GPT-4o performing best (MAE = 0.885). However, none of the models produce statistically significant overall rank correlations with human experts (Kendall's tau). Vocabulary analysis of generated rationales reveals characteristic evaluative biases in each model, and RAG is shown to improve rank correlation while worsening absolute accuracy, constituting an important negative result for text-based rule injection.

Medical image segmentation is essential for modern computer-aided medicine. Recently, text-guided segmentation has shown promise by incorporating clinician-formulated textual reports as semantic guidance for image segmentation. These textual reports contain language descriptions about the appearance, location, and neighboring anatomy of segmentation targets, providing explicit guidance for target localization and delineation. Existing text-guided segmentation methods typically extract textual semantics implicitly through a pretrained text encoder and then integrate vision-language semantics via straightforward image-text feature fusion. However, these methods do not explicitly capture target-oriented information embedded in textual reports, particularly target location, and do not explore multi-level information fusion strategies beyond basic feature-level fusion, limiting the extraction and integration of critical textual semantics. In this study, we propose LoG, a localization-infused vision-language fusion framework for text-guided medical image segmentation. By jointly performing multi-scale target localization tasks, LoG explicitly captures target-oriented vision-language semantics and enables three-level localization-infused semantic fusion: (i) localization-guided feature fusion that directly infuses location-relevant semantics into visual features, (ii) localization-gated attention fusion that redirects multi-scale localization predictions to reinforce critical regions, and (iii) localization-constrained loss fusion that supervises segmentation based on spatial consistency with target localization. Extensive experiments on three well-established benchmark datasets, involving three medical imaging modalities with paired textual reports, demonstrate that LoG consistently outperforms state-of-the-art medical image segmentation methods.

Large Vision-Language Models (LVLMs) typically require processing hundreds to thousands of visual tokens, leading to substantial inference overhead. Existing visual token pruning methods either operate before the LLM using text-agnostic heuristics or prune inside the LLM at the cost of efficiency and noisy cross-modal attention. To address these limitations, we propose CRISP, a pre-LLM yet text-driven visual token pruning framework that preserves both instruction-relevant evidence and essential scene context. CRISP works in a two-stage pipeline: Stage 1 first identifies text-aligned visual tokens, and Stage 2 enhances contextual completeness through semantic diversity. Extensive experiments on LLaVA-1.5 and LLaVA-NeXT demonstrate that CRISP achieves superior performance retention under aggressive pruning ratios, maintaining up to 99.5% accuracy while reducing inference cost and latency by more than 2 times. CRISP serves as a practical solution for efficient LVLM inference, especially in resource-constrained scenarios.

Foundation models provide powerful representations for brain MRI analysis, but their predictions remain difficult to interpret in anatomically meaningful terms. Clinical assessment of brain MRI is commonly organized around anatomically defined structures and regional abnormalities, whereas conventional explanation methods typically produce voxel- or patch-level importance maps that do not explicitly quantify the contributions of individual brain regions. To address this mismatch, we propose RegionFM, an interpretable framework that integrates anatomical segmentation with brain MRI foundation-model embeddings. RegionFM first divides each MRI scan into anatomical regions and constructs a separate MRI volume for each region. A frozen foundation model then encodes each region into an embedding, and a region-additive logistic model combines these embeddings such that every anatomical region contributes an explicit scalar term to the final prediction. This formulation supports both subject-level and cohort-level analyses of regional contributions. We evaluate RegionFM on cognitive-impairment classification using embeddings from multiple pretrained brain MRI foundation models. The results show that RegionFM maintains performance comparable to less interpretable fine-tuning approaches while providing anatomically grounded explanations. Randomized embedding ablations yield near-chance performance, indicating that the predictions rely on meaningful structure captured by the foundation-model embeddings rather than simple feature statistics. Overall, RegionFM better aligns model explanations with anatomy-based clinical reasoning while maintaining competitive predictive performance.

Malaria diagnosis in endemic regions depends on species-level identification of Plasmodium parasites in thick blood smears, but deep learning detectors classify detections without providing morphological evidence for their predictions, limiting the ability of microscopists to audit those predictions at the case level. We present SGMCE (Segment-Grounded Morphological Concept Explanation), a post-hoc explanation framework that requires no additional training, no morphological annotations, and no labelled explanation data, yet produces per-detection natural-language explanations anchored in thick-smear morphology. For each detection, SGMCE extracts mask-guided crop thumbnails, computes fourteen handcrafted computer-vision morphological features (shape, colour, chromatin, haemozoin pigment) using adaptive within-mask thresholds, and queries GPT-4o with both visual evidence and computed measurements, conditioned on a thick-smear-specific knowledge base compiled from the World Health Organization bench aids. The primary output is a structured explanation identifying which morphological features support the detected species and why the competing species are excluded. Explanations are validated by four automatic metrics: Knowledge-Base Consistency (KBC), CV-Claim Faithfulness (CCF), Discriminativeness Score (DS), and LLM-as-Judge (LLMj). A sentence-level semantic scoring rule with species-aware negation filtering resolves the vocabulary mismatch between clinical prose and knowledge-base terms. Across 737 detections from 139 thick-smear images spanning four Plasmodium species and white blood cells, parasite-class mean KBC is 0.91, mean DS is 0.99, and mean CCF is 0.97, while a per-rule CCF breakdown confirms that the CV-grounded claims made by the vision-language model are consistent with the measurements they cite.

Electrocardiography (ECG) is an inexpensive, standard-of-care test for cardiac symptoms, but front-line triage often lacks immediate access to definitive imaging such as echocardiography (ECHO) or cardiac magnetic resonance (CMR). Furthermore, most existing ECGAI systems are limited to fixed diagnostic labels or automated reports, constraining their use for patient-specific clinical reasoning. To address this gap, we introduce ECG-LLM, an ECG-conditioned large language model trained across four cohorts comprising 679,112 ECG studies from 186,409 patients. Using a novel multimodal-to-language supervision strategy, ECG-LLM is trained on clinically structured question-answer pairs derived from ECG signals, clinical context, CMR, and ECHO. This unified approach enables the model to answer diverse cardiovascular questions from a 12-lead ECG alone, spanning both conventional interpretation and phenotypes not directly visible on standard ECGs. ECG-LLM successfully recovers conventional ECG measurements, such as heart rate, and strongly predicts complex CMR-derived phenotypes, including ventricular and atrial volumes and ventricular function. Crucially, it detects vital echocardiographic phenotypes, including increased LV wall thickness, aortic stenosis, and right-ventricular systolic dysfunction. On standard ECG understanding tasks, ECG-LLM matches or exceeds existing baselines for diagnostic report generation and the ECG-QA benchmark. By moving beyond fixed-label prediction, this multimodal framework provides clinically valuable, question-driven cardiovascular reasoning to support general practitioner and front-line triage decisions when specialist review is delayed.

Micro-gesture recognition demands the detection of fleeting, spatially localized movements that are frequently overwhelmed by dominant static appearances and background noise. While Multimodal Large Language Models (MLLMs) excel at general video understanding, they inherently struggle with subtle kinematics and often rely on static posture priors. To this end, we propose GMoT, a Gated Motion-Aware Tokenization module that explicitly distills sparse kinematic evidence into a compact sequence prior to temporal modeling. GMoT dynamically spotlights action-relevant regions via spatially weighted pooling, extracts adjacent-frame temporal differencing to capture precise motion energy, and adaptively fuses these cues into the visual stream using a conservatively initialized semantic gate. To transition from simple classification to evidence-grounded reasoning, we further introduce a progressive reward-guided policy refinement paradigm, supported by a semi-supervised annotation pipeline that generates anatomically focused captions. Beyond achieving the best Top-1 accuracy among the compared methods on iMiGUE (67.32%) and SMG (73.11%), improving the Qwen3-VL-8B baseline by +6.80 and +3.11 points, our framework introduces Body-Region Grounding (BRG) Recall as an anatomical-grounding proxy conditioned on correct predictions, together with an overlapping-label cross-domain transfer protocol between iMiGUE and SMG. Extensive evaluations demonstrate that our GMoT-augmented model improves in-domain accuracy, retains clear gains under label-preserving corruptions, and improves accuracy-oriented cross-domain transfer under explicit small-split caveats while maintaining high anatomical grounding in its generated rationales.

Understanding a painting is never a single act. Art historians may analyze the same work through concepts of style, iconography, or historical context, dimensions that are not interchangeable, and each carries distinct semantic relationships between the visual and the textual. Vision-Language Models (VLMs) like CLIP, which learn a single shared embedding space, collapse this richness into a single homogeneous alignment, thereby losing the multi-relational structure that defines art-historical reasoning. We introduce CANVAS (Contrastive Art-aware Network for Vision-Language Alignment with Sheaves), a framework for learning relation-aware multimodal representations inspired by sheaf theory. Each artwork is projected into multiple embeddings conditioned on the type of relation (i.e., the context), and a novel contrastive loss encodes contextual information during training, with no dependency on external data at inference. We evaluate on three newly introduced benchmarks of artworks for multi-relational art understanding: WikiArt+, derived from WikiArt and Wikipedia, HertzianaDP, from the Bibliotheca Hertziana collection, and SemArt+, refined from the SemArt dataset. In multimodal retrieval and art understanding, CANVAS outperforms the baselines, supporting the view that multi-relational alignment is not just theoretically motivated but also practically essential.

In this paper, we reveal an important yet overlooked problem in image denoising: under signal-dependent camera noise models, dark regions suffer from inherently low Signal-to-Noise Ratio (SNR), as signal intensity decays far faster than noise variance diminishes, making detail recovery in dark areas fundamentally challenging. Yet rather than compensating for this difficulty, MSE-trained denoisers exacerbate it – reconstructing dark pixels up to 6x worse relative to their per-band noise floor. This bias stems from two compounding factors: signal-dependent noise inflates bright-pixel residuals, and the network's Jacobian norm increases monotonically with brightness. Together, these cause bright regions to chronically dominate gradient updates at the expense of dark ones. To this end, we propose Brightness Bias-Robust Denoising (BBRD), a drop-in replacement for MSE loss that partitions pixels into brightness bands, normalizes per-band error by empirical noise variance, and applies Group Distributionally Robust Optimization (Group-DRO) to dynamically upweight whichever band is currently worst, with zero additional parameters or inference cost. Across 8 architectures and 2 datasets in our experiments, BBRD is the only method among 13 tested alternatives that improves each brightness band simultaneously, achieving up to +0.45 dB on dark bands, +0.32 dB on bright bands, and +0.65 dB aggregate Peak Signal-to-Noise Ratio (PSNR) on SIDD, with the largest per-band gains in the darkest regions where detail recovery matters most. Code is available at https://github.com/xmed-lab/BBRD

As Ultra-High-Definition (UHD) displays and immersive media services become ubiquitous in the Internet of Things (IoT) and Consumer Electronics (CE) sectors, including 8K display and mobile devices, the demand for high-efficiency video coding is unprecedented. While Deep Learning-based Filtering (DLF) has emerged as a promising solution to mitigate compression artifacts inherent in standards like High Efficiency Video Coding (HEVC/H.265) and Versatile Video Coding (VVC/H.266), its deployment in CE devices is severely constrained by computational complexity, memory bandwidth, and power consumption. To bridge the gap between academic research and practical deployment, this paper presents a comprehensive, hardware-oriented survey of DLF techniques. We propose a systematic three-dimensional taxonomy classifying methods into (1) Integration Scheme within the Video Coding, (2) Coding Information Utilization, and (3) Network Design Strategy. Unlike prior reviews, this work critically analyzes the trade-offs between Rate-Distortion (RD) performance and hardware feasibility, highlighting the evolution from heavy, performance-oriented models to lightweight, hardware-friendly architectures targeting Neural Processing Units (NPUs). Furthermore, we incorporate the latest standardization activities from the Joint Video Experts Team (JVET) on Neural Network-based Video Coding (NNVC) to provide realistic guidelines. We also identify open challenges such as real-time inference latency and error propagation, providing a roadmap toward robust, low-power intelligent video coding in next-generation CE vision endpoints.

Deep networks now subtype brain tumors on MRI about as well as specialist readers, yet accuracy is not what keeps them out of the clinic. What matters at the point of care is whether a model's confidence can be trusted to flag the cases it is likely to misclassify and defer them to a human. Deterministic estimates cannot: an auxiliary confidence head trained alongside the classifier collapses to a near-constant output that says nothing about correctness. This study proposes an uncertainty-first pipeline for four-class brain tumor MRI (glioma, meningioma, pituitary, no tumor) that reads predictive uncertainty from Monte Carlo (MC) Dropout over T = 20 passes and turns the resulting entropy into an explicit rule for deferring uncertain cases to a radiologist. We partitioned 7,200 images by perceptual-hash cluster, closing the near-duplicate leakage that inflates accuracy under naive splitting, and evaluated the pipeline on ViT-B/16 and ResNet-50 across five seeds along three axes: discrimination, calibration, and selective prediction. Both discriminate strongly (macro-AUC 0.994; accuracy 0.962 and 0.964), and no seed separates them (0 of 5 significant, p < 0.05), so the result is driven by the uncertainty pipeline, not the network. A single temperature scalar pulls the deterministic softmax into tight calibration (expected calibration error 0.016-0.020), and deferring the most uncertain 5% of cases lifts accuracy on the rest to about 0.98 on both (area under the risk-coverage curve 0.010-0.011). MC-Dropout uncertainty here is thus calibrated, non-collapsing, and directly actionable through a concrete deferral rule, providing an architecture-agnostic basis for calibrated, defer-to-human brain tumor MRI triage under internal validation.

In this report, we introduce Eddy-VL 1.9B, a compressed multimodal embedding model built on Qwen3-VL-Embedding-2B for offline, edge-deployable vision-language retrieval. Eddy-VL targets air-gapped forensic and investigative settings where cloud APIs are unavailable and low latency is essential. Compression combines (i) probe-driven structural pruning that removes four redundant text-decoder layers (28 to 24) ranked by adjacent-layer linear CKA, and (ii) layered knowledge distillation with hole-covering teacher-student mappings, mid-layer attention-map 1-CKA, and final-layer MSE and cosine losses with Matryoshka dimensions {128, 256, 512, 1024, 2048}. The released model contains 1,926,188,032 parameters (3.85 GB bf16), representing approximately 9.5% fewer parameters than the 2.13B teacher model. Empirical evaluations on MMEB-V2 (78 tasks, VLM2Vec protocol) show that Eddy-VL achieves an overall score of 63.2 compared with 68.9 for the teacher, retaining 91.7% of the teacher's performance while recovering 6.4 of the 12.1 points lost through pruning alone (56.8). Compositional reasoning performance remains close to the teacher on SugarCrepe (86.1 vs. 86.4), MR2-Bench (24.5 vs. 24.7), and ARO (59.5 vs. 60.4), while Winoground group performance (6.8 vs. 8.5) remains the primary limitation. Depth pruning also reduces forward latency by approximately 10% (150.0 to 136.4 ms per image on NVIDIA DGX Spark using FlashAttention-2). We present the architecture, compression methodology, training procedures, and evaluation results, demonstrating the effectiveness of Eddy-VL for multimodal retrieval under constrained edge deployment. Model weights and inference code are publicly available on Hugging Face.

World models, especially based on JEPA architectures, have been shown to learn robust dynamics of various environments. However, learning from visually complex real-world data remains a challenge, especially in unpredictable outdoor environments. We introduce depth as a geometric prior during training in learning more robust latent dynamics directly from robot video data and handling visual complexity. This combines depth supervision with an isotropy-inducing latent regularizer (SIGReg), maximizing task-agnostic latent diversity while constraining how that diversity is organized, with the combined objective targeting the highest-entropy representation consistent with scene geometry. To satisfy this greater complexity without increasing inference time, we also add training-only overparameterization. Training an 18M-parameter model on video from a real agricultural robot, we evaluate with frozen-representation visual odometry probes, predictor-based surprise detection, and multi-step latent rollout fidelity. Compared to the baseline LeWM, our method lowers visual odometry probe error by 33%, substantially increases surprise-score separation both in-domain and on the out-of-domain TartanGround benchmark, and improves multi-step rollout fidelity under domain shift, with gains that grow with rollout horizon. Notably, we also see improvements in surprise-score separation on physics understanding that is not directly tied to 3D geometry, such as lighting and shadows. These results show that a lightweight training-time geometric prior makes a compact JEPA world model more useful and more transferable on real outdoor data with strong underlying representations, without adding inference overhead. Our work suggests that depth as a physically grounded prior can enhance world model generalization on a variety of tasks.

Frozen self-supervised vision models can align parts of generic objects, but it remains unclear whether this correspondence extends to human faces, where global layout is shared while identity-specific appearance varies sharply. We test whether frozen DINOv3 features define a region-level facial coordinate system: a feature space in which eyes, brows, nose, mouth, skin, and hair remain distinguishable across people and across time without face-specific training. Using DINOv3 ViT-L/16 patch embeddings and FaRL only as a face-part labeling interface, we evaluate cross-identity nearest-neighbor matching and temporal label propagation on 200 CelebDF-v2 real videos. DINOv3 achieves 83.0% region-level semantic accuracy under unconstrained cross-identity matching, compared with a 23.0% area-weighted random baseline, and 95.5% temporal tracking accuracy without a learned temporal module. A no-FaRL control collapses to 0.9%, showing that FaRL supplies semantic initialization while DINOv3 supplies dense spatial correspondence. The strongest correspondence appears at an intermediate layer: block 18 gives a 4.93x same-region versus cross-region discrimination ratio, compared with 1.48x at the final block. Against CLIP ViT-L/14, DINOv3 shows only a small aggregate advantage but a +16.8 pp gain on anatomical regions, indicating that image-level contrastive supervision captures coarse facial layout but not fine-grained anatomical identity. These results establish frozen DINOv3 as a strong zero-shot representation for region-level facial correspondence and identify intermediate self-supervised features as the most useful layer for dense face analysis.

Computed Tomography (CT) carries significant ionizing radiation risks, driving the need for sparse-view reconstruction. Implicit scene representations (ISRs) address this by recovering continuous volumetric attenuation fields directly from sparse projections, and recent geometry-aware extensions jointly model surface geometry alongside attenuation to improve fidelity and enable clean tissue segmentation without manual thresholding. However, these methods remain limited by manually tuned attenuation bounds and rigid two-material constraints. This paper proposes $K$-NeAS, a unified and scalable architecture for automated, multi-material surface reconstruction. We replace independent material networks with a shared latent backbone and introduce a fully differentiable $K$-material sequential soft selector to model an arbitrary number of overlapping tissues. To eliminate manual tuning, we automate attenuation bounding using a Gaussian Mixture Model (GMM) and implement a scheduled auxiliary floater loss to mitigate geometric hallucinations common under extreme sparsity. Evaluated across four clinical Cone-Beam CT (CBCT) datasets, $K$-NeAS successfully scales to arbitrary material counts, achieving superior 3D volumetric fidelity at $K=3$ materials on complex multi-tissue regions such as the Abdomen ($33.28\text{ dB}$ 3D PSNR vs. $31.40\text{ dB}$ single-material NeAS baseline, a $+1.88\text{ dB}$ improvement). Furthermore, our model exhibits enhanced robustness under sparse-sampling conditions, outperforming baseline 3D PSNR by up to $1.17\text{ dB}$ under 5- and 10-view constraints.

Fine-grained manipulation recognition requires modeling evolving relations among hands, objects, tools, and supporting surfaces. Conventional graph-based methods use pairwise edges that can fragment a coordinated event into disconnected binary relations. We propose a dynamic manipulation hypergraph framework that represents multi-entity configurations as higher-order relational units. At each temporal step, relevant entities are encoded using appearance, spatial, motion, and semantic-role features. Hyperedge candidates are instantiated and ranked using proximity, contact, and motion-coupling predicates. A hypergraph reasoning network performs node-to-hyperedge and hyperedge-to-node message passing, followed by temporal attention over the evolving interaction structure. The framework provides class-agnostic hyperedge-importance scores that identify entity configurations and temporal intervals emphasized by the model without treating them as causal explanations. Quantitative evaluation is conducted on EPIC-KITCHENS-100/VISOR and Assembly101 under an annotation-assisted entity-localization protocol. Video-only and entity-based methods provide contextual comparisons, while a matched pairwise graph and a static hypergraph serve as the principal controlled baselines because they use identical entity inputs and comparable relational settings. The proposed method improves HO-F1 over the matched pairwise graph by 6.9 percentage points on EPIC-KITCHENS-100/VISOR and 9.5 points on Assembly101, and exceeds the static hypergraph by 4.4 and 5.8 points, respectively. Qualitative analysis on ARCTIC further shows correspondence between highly ranked hyperedges and contact-rich manipulation intervals. These results demonstrate the value of time-varying higher-order relational modeling for fine-grained manipulation activity recognition.

Region-based loss functions, such as the Dice loss, have established themselves as the de facto standard for highly class- and region-imbalanced segmentation tasks. However, models trained using region-based loss functions are notoriously miscalibrated and typically yield over-confident predictions. In medical imaging applications, such as defining tumor resection margins, this miscalibration is hindering clinical adoption. In this work, we outline a novel gradient perspective on this overconfidence and show how it affects region-based loss functions. We propose a "surgery" on the gradient vector field as a simple, yet effective intervention to mitigate calibration issues. This surgery adds a factor to the loss's partial derivative, scaling the gradient's magnitude linearly with the prediction error. In empirical evaluations across 2D and 3D medical segmentation tasks, we demonstrate the effectiveness of this intervention while maintaining high prediction accuracy when used in conjunction with any region-based loss function.

Learned image compression (LIC) is bottlenecked by the need to store independent models for each rate-distortion operating point. Existing variable bit-rate (VBR) methods aim to reduce this overhead via dense parameter modulation, but forcing a shared backbone to approximate divergent mappings causes severe feature entanglement. Specifically, low-rate smoothing gradients inherently conflict with the preservation of high-frequency textural details, leading to sub-optimal performance. To resolve this, we propose MixCompress, a unified VBR framework based on sparse structural specialization. While sparsely gated Mixture-of-Experts (MoE) routing successfully mitigates gradient conflict, it operates on a fixed computational budget. To address the increased representational demands of higher bit-rates we introduce a Mixture-of-Depths (MoD) extension to dynamically scale model capacity. Combined with Conditional Auxiliary Transforms (CAT) for dynamic sub-band energy modulation, our hierarchical framework effectively dynamically scales capacity. Extensive evaluations demonstrate that MixCompress not only matches individually optimized single-rate baselines but can even surpass them, establishing a new Pareto frontier for computationally efficient image coding.

Vision Language Models (VLMs) demonstrate strong perceptual abilities but remain limited in tasks requiring analytical reasoning across multiple visual states, such as multi-image comparison, change detection, and multi-step visual inference. These capabilities are critical for real-world multimodal applications where reasoning must be grounded in systematic differences between visual contexts. However, existing benchmarks rarely require both explicit visual comparison and analytical reasoning, leaving this capability underexplored. To address this gap, we introduce SD-MAR (Synthetic Data for Multi-image Analytical Reasoning), a framework for training and evaluating VLMs on multi-image analytical reasoning. SD-MAR constructs paired visual scenarios through controlled perturbations and generates reasoning tasks spanning semantic change attribution and quantitative comparison. We further train VLMs using GRPO-lite with Backward Discounted Allocation (BDA), a reinforcement learning approach that removes KL regularization to encourage stronger policy optimization while allocating greater credit to the later reasoning steps where analytical conclusions are formed. Experiments on Qwen2.5-VL-7B and InternVL3-8B show that GRPO-lite fine-tuning on SD-MAR improves in-domain accuracy by up to 36.95%, with Qwen2.5-VL-7B outperforming GPT-4.1 on the SD-MAR benchmark. Importantly, out-of-domain generalization is preserved or improved: performance remains within 1% on MME, MMMU-Pro, and MathVista, while improving by up to 4% on MMBench. LLM-as-judge evaluation further demonstrates consistent improvements in logical coherence and explanation quality across both models.

In proton therapy planning, respiratory-gated non-contrast CT (NCCT) is commonly used for lesion segmentation; however, accurate delineation remains challenging due to low lesion-to-background contrast. Although learning-based methods have shown strong performance, they often struggle with non-contrast image segmentation. Inspired by clinical practice, where contrast-enhanced MRI is referenced to delineate lesions on NCCT, we propose ViPSAM, a visual prompting framework that leverages complementary cross-modality information. Built upon the Segment Anything Model (SAM), ViPSAM introduces a visual prompt encoder to extract guidance features from contrast-enhanced images and a visual-guided cross-attention module to integrate non-contrast and contrast-enhanced features, thereby enhancing lesion-relevant representations in low-contrast regions. The mask decoder is further adapted in a parameter-efficient manner to utilize visual prompts effectively. We evaluate the proposed method on liver lesion segmentation using NCCT acquired for proton therapy. Experimental results demonstrate that ViPSAM outperforms representative U-Net- and SAM-based methods, indicating that cross-modality visual prompting enables more robust and accurate segmentation in non-contrast images.

In this paper we propose DCVC-Mamba (DCVC-MB), a neural video codec framework for B-frame coding. Our approach incorporates an IBP frame strategy for low-delay B-frame coding, a spatio-temporal fusion model based on state-space models for bidirectional temporal prediction, and an entropy-aware skipping mechanism that selectively omits coding certain latents to reduce entropy coding times. In addition to our model contributions we also implement two inference-time strategies that enhance compression performance. Experimental evaluation shows that DCVC-MB compares favorably to existing NVCs and traditional codecs. The method demonstrates BD-rate reductions of up to $8.98\%$ on average compared to prior neural video codecs, and improvements of up to $30.45\%$ and $1.81\%$ over the VTM-19.0-LDP and VTM-19.0-RA(Inter-GoP=16) benchmarks, respectively, contributing to advances in neural video compression.