AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Spatio-Temporal Conditional Denoising Transformer for Modality-Missing RGBT Tracking

Andong Lu, Ziyi Zha, Jiandong Jin, Shihao Li, Chenglong Li, Jin Tang, Bin Luo

Missing modalities in RGBT tracking often lead to incomplete and unstable multimodal feature representations that greatly degrade the performance. Existing methods typically attempt to recover missing modalities from available ones, but the quality of data generated in challenging scenarios might be unsatisfactory. In addition, current approaches exhibit limited flexibility in processing both missing and complete data. To overcome these limitations, we propose a Spatio-temporal Conditional Denoising Transformer (SCDT), which integrates the spatial cues and the temporal context to adaptively perform information reconstruction of missing modalities and feature enhancement of weak modalities in a unified framework, for robust modality-missing RGBT tracking. In particular, SCDT leverages the short-term temporal cues from recent historical frames to capture the fine-grained temporal correlations and the long-term temporal cues encoding modality evolution to capture the global context. By jointly exploiting long short-term temporal contexts as the conditions, SCDT progressively guides noisy features of available modalities to learn reliable and temporally consistent multimodal representations. Furthermore, SCDT introduces a noisemodulated adaptation mechanism that dynamically adjusts its behavior according to the modal availability, enabling a single framework to unify feature learning under both modality-missing and complete scenarios without changing the architecture or parameters. Extensive experiments on three public benchmark datasets demonstrate that our method consistently outperforms state-of-the-art methods. The code is available here.

arXiv

Co-Learning for Missing Arbitrary Modalities in Multi-modal Classification

Francisco Mena, Dino Ienco, Roberto Interdonato, Cassio F. Dantas, Simon Besnard

Multi-modal classification leverages complementary information across diverse data sources to enhance predictive performance. However, real-world scenarios subject to operational constraints, such as sensor failures or privacy restrictions, lead to inconsistent modality availability between training and inference times. To handle missing modalities, prior studies have mainly covered bimodal data setups and focused on designing robust fusion processes. Instead, we adopt a multi-modal co-learning framework that prioritizes inter-modal collaboration rather than multi-modal fusion. Specifically, we consider that any subset of modalities may be absent, without assuming predefined missing-modality patterns, an inference scenario we refer to as missing arbitrary modalities. To address this challenge, we introduce two alternative approaches that leverage information at both feature- and decision-level. Experiments on two multi-modal classification benchmarks demonstrate significant robustness gains in various missing modality conditions. The first method shows more robust behavior under minimal missing conditions, where a single modality is absent, whereas the second performs better under extreme missing conditions, where all-but-one modalities are missing. Our code is available at https://github.com/fmenat/Co4Miss.

arXiv

MMOE: Modernizing Diffusion Transformers with Efficient Expert Design

Yanhao Jia, Jiepeng Wang, Haibin Huang, Chi Zhang, Erik Cambria, Xuelong Li

Modern large language models scale successfully by pairing capacity growth with efficiency, keeping per-token and deployment costs under control as capacity grows. AIGC Foundation Models (AFMs), especially diffusion-transformer backbones, have begun to adopt sparse experts, but recent efforts mostly enlarge total parameter counts and sparsity ratios without importing the efficiency mechanisms that made LLM scaling practical, so generation quality is seldom balanced against training and deployment cost. This raises a natural question: can the architectural principles behind efficient LLM scaling be adapted to AFMs in a more balanced way? We introduce ModernMOE (MMOE), a modernization of SiT-style diffusion transformers that systematically adapts routed experts, shared and lightweight experts, gate-residual routing, and attention-residual information reuse to AIGC generation. Rather than treating MoE as a single plug-in replacement, MMOE studies how different modern expert components affect convergence, efficiency, and generation quality when composed inside a diffusion transformer. Every experiment in this paper is trained on a single eight-GPU H100 node with batch size 256 for 400k steps, an accessible single-machine budget. Under matched training and sampling protocols and at this budget, MMOE reaches lower FID at every recorded checkpoint, that is, it converges faster per training step, than dense and intermediate sparse-expert baselines, and among the sparse variants it attains the best quality-cost balance. Routing analysis further shows stable expert specialization across depth, substantial use of lightweight routes, and modest step-to-step routing changes during denoising. These results suggest that AFMs can follow the balanced scaling path of LLMs by importing proven efficiency designs, rather than by simply increasing total parameters and sparsity ratios.

arXiv

Test-Time Adaptation via Dual Distillation for Videos Under Severe Distribution Shifts

André Sacilotti, Samuel Felipe dos Santos, Jurandy Almeida

Deep learning models have achieved state-of-the-art performance in several computer vision tasks. However, they experience severe performance degradation when applied to real-world scenarios due to unanticipated distribution shifts. Test-Time Adaptation (TTA) attempts to solve this problem by using unlabeled data from the target domain to dynamically adapt to the test distribution at inference time, without access to the source data. However, TTA remains a challenging problem when adapting to continuous, temporally correlated data, such as videos, and in scenarios where the target domain contains severe domain shifts. For this reason, few works in the literature explore TTA for videos under such extreme conditions. To overcome these limitations, we propose Test-time Adaptation via Dual Distillation (TADD), an online adaptation framework that relies on a lightweight projection adapter to bridge the domain gap. The adapter module is pre-trained on the source domain and then adapted to the target using our proposed complementary losses: (i) zero-shot distillation, which encourages alignment with the domain-agnostic features from a pre-trained vision-language model (VLM); and (ii) target distillation, which retains the source domain discriminative knowledge encoded in the pre-trained adapter. Built upon a frozen CLIP backbone, our method introduces this lightweight projection adapter as the sole updatable component during inference. We conducted extensive evaluations on three well-known video action recognition benchmarks: UCF-HMDB, Daily-DA, and Sports-DA. Our experiments in the closed-set scenario demonstrate that our method consistently outperforms state-of-the-art TTA baselines. Notably, our TTA approach improves upon previous methods by up to +3.81% on UCF-HMDB, +2.63% on Daily-DA, and +3.03% on Sports-DA.

arXiv

QueenVIS: Rethinking Image-Only Training for Video Instance Segmentation via Query Enrichment

Arian Kheirandish, Fardin Ayar, Ehsan Javanmardi, Manabu Tsukada, Mahdi Javanmardi

Video instance segmentation (VIS) requires models to detect, segment, and track object identities across frames, and most methods enforce temporal consistency through video-level supervision. Image-only training approaches, with MinVIS as one prominent example, have challenged this assumption, reaching competitive VIS without video training by treating frames as independent images and associating instances only at inference. The field has nonetheless moved toward ever more elaborate video-trained trackers, which depend on costly identity-consistent annotations, leaving the image-only direction under-explored. A diagnostic analysis identifies object query quality as the bottleneck: queries trained only to localize objects within a frame drift apart across frames and destabilize association. QueenVIS introduces a query-centric framework for strengthening image-trained VIS. During single-frame training, we enrich Mask2Former queries with two auxiliary heads: a feature-prediction loss that aligns each query with the pooled backbone descriptor of its instance, and a center-prediction loss that injects spatial structure. Both heads are discarded at inference, adding zero parameters, and temporal identity is maintained by a training-free query-propagation and memory-bank scheme. On YouTube-VIS and OVIS with a ResNet-50 backbone, QueenVIS improves over MinVIS, up to +6.7 AP on YouTube-VIS, +4.8 AP on OVIS, and +10.3 AP on the long-sequence YouTube-VIS split. QueenVIS achieves 50.9 AP on YouTube-VIS and remains competitive with recent video-supervised state-of-the-art, without processing a single video clip during training. Our findings suggest that strengthening the discriminative power and temporal stability of object queries is an important, underexplored axis for VIS. Code and models: https://github.com/ArianKheir/QueenVIS

arXiv

CameraAnything: Refilming Videos with Arbitrary Camera Control

Yixuan Li, Yanhong Zeng, Ka Leong Cheng, Jiayi Zhu, Hanlin Wang, Wen Wang, Yihao Meng, Hao OUYANG, Qiuyu Wang, Yue Yu, Zidong Wang, Yiyuan Zhang, Yujun Shen, Dahua Lin

We introduce CameraAnything, the first unified framework for camera controlled video editing that enables joint control of both intrinsic and extrinsic camera parameters. Existing approaches either rely on expensive 3D reconstruction to achieve full camera functionality or restrict editing to extrinsic parameter manipulation. Moreover, the coupled influence of intrinsic and extrinsic parameters on video appearance makes disentangled modeling particularly challenging. To address this, we adopt per-pixel Plücker ray injection alongside resolution-aware 3D RoPE in self-attention, building both camera conditioning and spatial positional encoding on the target latent to jointly control camera position, focal length, and native resolution editing without cropping or outpainting. To overcome the scarcity of paired training data, we further develop a scalable synthetic pipeline that constructs diverse dynamic scenes through structured multi-camera recording and generates synchronized videos with varied camera configurations. With a tailored orthogonal training strategy, CameraAnything enables expressive video reshooting with arbitrary viewpoint control, focal length adjustment, resolution adaptation, and multi-shot transitions within a single generation process, offering strong practical value for cinematic video editing and cross-platform content adaptation in video production.

arXiv

CADER: Confidence-Aware Dynamic Evidence Reasoning for Long-Video Understanding

Jinlong Yang, Wenhao Zhang, Kuanwei Lin, Sijie Cheng

Long-video understanding increasingly relies on large vision-language models and tool-augmented reasoning, but most systems apply the same inference procedure to every example regardless of difficulty. This uniform strategy invokes unnecessary tool-assisted processing for easy questions and provides limited control when difficult questions require fine-grained temporal evidence. We propose CADER (Confidence-Aware Dynamic Evidence Reasoning), a training-free framework for adaptive and reliable long-video reasoning. CADER first performs global reasoning over uniformly sampled frames and estimates answer confidence with a logit-margin signal, allowing high-confidence examples to exit early. For uncertain examples, CADER activates a second-stage tool-augmented loop that combines temporal cropping, lightweight semantic verification, and Relevance-Guided Resampling to progressively localize question-relevant evidence. This design treats tool use as a sample-level decision: a single global pass handles easy cases, while additional reasoning is reserved for examples where uncertainty suggests that more evidence is needed. Experiments on multiple VideoQA benchmarks show that CADER improves long-video reasoning while bypassing Stage~2 for high-confidence samples. Moreover, when applied to a backbone trained only with tool-free chain-of-thought supervision, CADER achieves competitive performance against specialized tool-augmented frameworks, suggesting a practical inference-time route for adaptive long-video reasoning.

arXiv

Denoising 3D images: robustness of persistent homology measures

Ebru Dagdelen, Aakash Karlekar, Manav Arora, Matthew Illingsworth, Jonathan Jaquette, Linda J. Cummings, Lou Kondic

When computing sub/super-level-set persistent homology (PH), the effect of noise may introduce millions of (short-lived) topological generators, presenting an obstacle to both the computation of PH of large 3D images, and any analysis of PH that incorporates the number of generators. As such, it is often necessary to denoise the data before computing its PH. We analyze the PH of synthetic 3D images of porous media in the presence of spatially uncorrelated noise, and perform a comparative analysis of various topological measures (e.g. bottleneck distance, Wasserstein distance, persistence statistics and persistence images) to assess their robustness to both noise and the denoising process (i.e. adding spatially uncorrelated Gaussian noise, and denoising by either a Gaussian convolution or a machine learning approach).

arXiv

The Visual Bottleneck: Sparse-Frame Adaptation of MLLMs for Joint Spatial-Temporal Video Grounding

Jiameng Zhang, Srikanth Madikeri

Large-scale video platforms process millions of uploads hourly, requiring moderation systems that can localize when and where policy violations occur within each video. Processing every frame is infeasible at scale, so systems are constrained to sparse inputs of 8 to 16 frames per video. Yet state-of-the-art multimodal large language models (MLLMs) are pretrained on dense sequences of hundreds of frames, creating a fundamental mismatch between training and deployment conditions. This mismatch causes severe performance collapse: the Qwen3-VL 8B model drops from 56.0% to 22.3% temporal mIoU when frames are reduced to 16, a 60.2% relative degradation. We present a systematic empirical study of training strategies to close this gap for spatial-temporal video grounding. Our results suggest that visual feature extraction is the dominant bottleneck under sparse-frame inputs. Adapting only the final three ViT layers, 4% of total parameters, achieves 68.8% temporal mIoU and surpasses a zero-shot 8B model using dense inputs by 12.8 points. Language model fine-tuning, by contrast, offers negligible or negative returns. A boundary-aware sampling strategy, Hybrid16, further improves temporal mIoU by 26 points over uniform sampling when temporal boundaries are available. We conclude that for sparse-frame video grounding, training strategy dominates model scale: a fine-tuned 2B model consistently outperforms a zero-shot 8B model, with or without dense frame access.

arXiv

DSCH-Loss: A Dynamic Semantic Channel Objective for Deep Semantic Hashing

Tobias J. Bauer, Christian Riess, Daniel Loebenberger, Christian Bergler

Semantic hashing methods for generating short binary hash codes that allow efficient approximate nearest neighbor search in high-dimensional data spaces have gained extensive consideration in recent years. Deep learning-based methods offer better semantic capturing capabilities than traditional approaches relying on manual feature engineering. Moreover, they enable a data-driven approach to semantic hashing across diverse data modalities, yielding high-quality cross-modal hash codes within a shared Hamming space. Previous work investigated the properties of this Hamming space and introduced a loss function based on predefined so-called semantic channels with fixed width and Hamming distances derived from label similarities. However, this formulation also introduced discontinuities into the loss landscape, complicating optimization. Based on these observations, we propose a newly designed loss function, Dynamic Semantic Channel Hashing (DSCH), using dynamically sized and positioned semantic channels in order to avoid loss landscape discontinuities. Furthermore, we endorse the use of tie-aware Mean Average Precision (mAP) as evaluation metric as it addresses the ambiguity in sample retrieval ordering, which emerges from the discreteness of hash code distances. Finally, multiple experimental settings conducted on two popular datasets and incorporating two different model architectures provide strong evidence that training using the DSCH objective outperforms training using other state-of-the-art loss functions. In a total of 35 out of 40 cross-modal and intra-modal retrieval tasks, models trained with DSCH achieve significantly higher tie-aware mAP scores across all four tested hash code lengths, showing compelling results across model architecture and used dataset. The mAP score uplifts are consistent and amount up to 1.75 percentage points compared to the respective second best.

arXiv

EgoPlay: Event-Triggered Video Editing for Egocentric Streams

Jinjie Mai, Gordon Guocheng Qian, Willi Menapace, Arpit Sahni, Chaoyang Wang, Ashkan Mirzaei, Runjia Li, Sergey Tulyakov, Bernard Ghanem, Peter Wonka, Rameen Abdal

We introduce EgoPlay, an event-triggered video-to-video editor for egocentric streams, obtained by fine-tuning a pretrained V2V diffusion transformer on event-conditioned data built primarily from Ego4D. Given a monocular video and an event-triggered prompt of the form "when X happens, do Y," EgoPlay infers whether and when event X occurs, preserves pre-event frames, and applies edit Y only to the post-event continuation. Rather than cascading a separate event detector with an editor, EgoPlay learns event recognition, temporal restraint, and pixel-level editing jointly in a single end-to-end model, while also handling negative and multi-event prompts. To support this, we construct a large-scale dataset of 106K event-triggered clip-prompt pairs spanning positive triggers, fabricated-trigger negatives, and multi-event prompts. We then train a bidirectional video diffusion editor with event-triggered supervision and derive a causal variant for chunk-by-chunk streamable inference. We further introduce an event-aware evaluation protocol that separately measures post-trigger editing quality, pre-trigger preservation, and false-trigger robustness. On the Ego4D benchmark, EgoPlay substantially outperforms EgoEdit, the state-of-the-art instruction-based egocentric video editing baseline, with relative gains of 17.7%, 16.9%, and 16.4% in editing quality, visual quality, and background consistency. It also surpasses a VLM-guided detector-editor baseline by 15.7%, 14.5%, and 13.5% on the same metrics, while using less than half the GPU memory.

arXiv

DeCoRAG: Cognitive Decoupling and Semantic-Aware Cropping for Complex Document Understanding

Shuo Wang, Kai Zhang, Wenyuan Huang, Yizheng Yu, Xia Liao, Junming Su, Qing Wang, Fang Xi

Advancing multimodal retrieval-augmented generation (RAG) for complex document understanding presents a formidable dual dilemma of accuracy and efficiency, particularly in graph RAG. Processing structurally sparse yet visually dense layouts, such as extracting a tiny data marker from a financial chart, often incurs computationally prohibitive token overhead while still triggering catastrophic hallucination. However, multimodal Graph RAG pipelines rely on graph-construction stages that assume Vision-Language Models (VLMs) can resolve sparse semantics within high-density layouts. We challenge this assumption, revealing that forcing VLMs to localize visual evidence, interpret semantics, and extract relations triggers a "Visual Attention Sink," a mechanism driving catastrophic semantic loss, while full-page processing incurs massive computational overhead. Controlled interventions verify that this failure is boundary-driven rather than content-specific and that semantic anchoring mitigates it. To fundamentally correct this flawed paradigm, we introduce DeCoRAG, a multimodal Graph RAG pipeline that shifts knowledge processing from coupled visual-semantic reasoning to "Cognitive Decoupling." Rather than passively processing raw pixels, its graph-construction stage establishes a macroscopic Semantic Anchor to neutralize the attention sink. This anchor subsequently drives our Region-Aware Pruning and Cropping (RAP-Crop) mechanism, shifting the reasoning space from dense, noisy backgrounds to purified, intent-driven semantic clusters. The resulting graph supports hybrid retrieval and answer generation. Across complex document benchmarks, DeCoRAG improves the semantic pass rate by up to 12.5 percentage points over the strongest baseline and generalizes to DocVQA. RAP-Crop reduces offline graph-construction prompt tokens by 40.8% without sacrificing end-to-end accuracy.

arXiv

A Modern ConvNet for Solar Filament Detection

J. R. Hu, Q. Hao, Z. Zheng, P. F. Chen, C. Li, Y. Meng

Automated solar filament detection using deep learning faces several challenges. Semantic segmentation of solar filaments is a complicated multiscale feature extraction task with long-tail distribution. Furthermore, a large-scale, highly complete, and finely detailed dataset has become mandatory for providing abundant information. To address these challenges, we present a series of machine learning approaches to develop a solar filament detection workflow that performs superbly. First, we manually annotated a small-scale solar filament dataset based on H$α$ spectra called MHAS. Next, we developed the Multiscale ORiented DENdritic (MORDEN) model, a semantic segmentation model focusing on multiscale feature extraction. We also introduced the Dense Conditional Random Field (DenseCRF) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) methods for post-processing. Using the proposed workflow, we generated a large-scale, high-quality dataset called AHAS. Experimental results demonstrate that MORDEN outperforms several existing solar filament semantic segmentation models with open access. DenseCRF has been demonstrated to effectively capture fine edge details. We also evaluated the effects of data scaling and the reliability of DBSCAN and found that both approaches yield satisfactory performance. Multiple visualization results substantiate our quantitative findings. Our work provides a foundation for maximizing the potential of deep learning models for solar filament detection.

arXiv

FlowCTS: On-policy Continuous Trajectory Supervision of Flow Models

Kaiyang Ye, Yuan Ge, Junxiang Zhang, Bei Li, Ziming Zhu, Haishu Zhao, Xiaoqian Liu, Chenglong Wang, Jingbo Zhu, Zhengtao Yu, Tong Xiao

While on-policy distillation (OPD) effectively addresses sparse rewards and exposure bias in large language model post-training, its extension to flow models remains underexplored. To this end, we propose Flow Continuous Trajectory Supervision (FlowCTS), which matches subsequent student and reference trajectories initialized from the same student-visited state. Using the integral relation between trajectories and velocity fields, we derive a temporally weighted velocity-matching upper bound and discretize it into practical objectives parameterized by the number of supervision steps. Under a multi-reference setup, single-state FlowCTS-OPD outperforms vanilla KL-based OPD with faster convergence. FlowCTS-OPD improves GenEval from 0.90 to 0.93, OCR from 0.90 to 0.92, and PickScore from 22.75 to 23.06, while outperforming a mixed-reward RL baseline across all target metrics. Further analysis reveals a clear temporal supervision mismatch in vanilla KL-based OPD arising from its auxiliary SDE transition kernels. Beyond on-policy setting,FlowCTS also consistently outperforms vanilla SFT , particularly on OCR, while increasing supervision steps exhibit a trade-off between richer trajectory information and greater optimization difficulty.

arXiv

DecoupleMix: Decoupled Ratio Search and Convex Allocation for Scalable VLM Data Recipes

Jiahao Xie, Zhongbin Guo, Qianle Wang, Ruiqi Lu, Dongling Xiao, Wanxuan Sun, Cheng Yang

While data curation for Vision Language Models (VLMs) is increasingly active, public practice for constructing pretraining mixtures remains largely heuristic: practitioners stack datasets that pass quality filters, set cross-domain ratios by intuition, and lack a principled, attributable criterion for admitting new data, while frontier recipes remain undisclosed. We formulate data construction as a systematic mixture-optimization problem and turn it into a reproducible engineering discipline by decoupling the mixture into two orthogonal sub-problems: inter-class ratios across capabilities and intra-class ratios within a category. For inter-class allocation, we use a single-variable iterative search; for intra-class composition, we apply a multidimensional, dataset-level assessment scoring Quality and Difficulty, and formulate selection as a constrained convex optimization with a diversity objective. The DecoupleMix framework delivers two critical capabilities: guiding what data to collect next and rendering dataset validation a controlled, attributable experiment. Experiments show our approach consistently surpasses heuristic baselines. Moreover, optimal ratios discovered on small-scale proxies transfer seamlessly to larger scales without retuning. Using 80B additional multimodal continue-pretraining tokens, our VLM is competitive with strong open-source models trained with substantially larger multimodal budgets.

arXiv

NSL-SLAM: High-Fidelity Neural Structured-Light Depth for Practical SLAM and Reconstruction

Jiaheng Li, Binsheng Zhang, Xinhai Chang, Wenzheng Chen

Structured-light (SL) cameras power depth sensing in millions of devices, and recent neural SL decoding methods have substantially improved their depth quality. SLAM systems can benefit greatly from such strong depth sensing, where reliable geometry enables stable tracking and faithful reconstruction. In this work, we present NSL-SLAM, a practical SLAM system tailored for high-fidelity structured-light depth. We first strengthen SL depth sensing: inspired by the neural structured-light (NSL) method, we further incorporate strong monocular depth priors into the SL stereo decoding, reducing depth RMSE by 35% on Replica-SL compared to NSL. We then build a depth-centric SLAM pipeline with this stronger depth: because structured-light geometry is dense and metrically accurate, we keep it as the primary tracking signal, and add only sparse visual correspondences for geometrically degenerate cases and lightweight bundle adjustment for long-range drift. Our depth estimator and SLAM design reinforce each other: stronger depth makes a simple SLAM pipeline effective, and the depth-centric pipeline ensures this advantage transfers to downstream reconstruction. Experimentally, on the synthetic Replica-SL benchmark, NSL-SLAM achieves the best tracking accuracy and improves reconstruction F-score by 1.6 points over the SOTA baseline under a shared-depth protocol. On a real benchmark of 8 challenging scenes, it is the only method that avoids catastrophic failure on all sequences while achieving 43.3% lower trajectory deviation than selected baselines. The SLAM system runs online at 20.9 FPS, demonstrating that stronger structured-light depth and depth-centric system design together enable practical, robust SLAM.

arXiv

Rethinking Expert Training for Model Merging with Prompt Learning

Christos Georgakilas, Aniello Panariello, Samir El Karrat Moreno, Simone Calderara, Dimosthenis Karatzas, Joost van de weijer

Model merging aims to combine multiple domain-specialized experts trained from a shared foundation model into a single multi-task model. Existing approaches largely focus on improving the merging procedure itself and typically assume experts obtained through full-parameter fine-tuning. In this work, we revisit expert training for model merging. We first show that prompt-based adaptation provides a strong baseline: independently learned prompts can be exploited across tasks while keeping the backbone fixed, avoiding the interference introduced by weight merging. Building on this observation, we introduce Dual-Tuned Experts (DTEs), a two-stage training strategy that first learns prompts and then fine-tunes the vision encoder. This reduces the magnitude of task-specific parameter updates and produces experts with higher merge compatibility. Experiments across multiple CLIP architectures, full fine-tuning, and LoRA experts show that DTEs consistently improve merged performance of standard merging approaches and remain effective even when combining heterogeneous sets of experts.

arXiv

ESRVS: Extreme Semi-Supervised Retinal Vessel Segmentation with a Single Annotated Image

Mingzhi Xu, Yizhe Zhang

Learning from minimal human supervision is a long-standing goal in medical image analysis, where dense expert annotations are costly. We study retinal vessel segmentation in an extreme semi-supervised setting with one annotated image and a pool of unlabeled images. We propose ESRVS, which selects a representative reference image for manual annotation and transfers vessel cues using target-domain-adapted DINOv3 features. ESRVS constructs a multi granular vessel prototype, combines prototype-similarity maps with a physics-inspired prior to generate initial pseudo-labels, and refines the transferred supervision through weighted pseudo-label training and adversarial refinement. Across eight public datasets, ESRVS achieves the best Dice and clDice on six datasets, and the best HD95 on all eight datasets among the compared semi-supervised methods, although those methods use 10 to 20% labeled data. With Mask2Former, ESRVS retains on average 93.7% of fully supervised Dice and 95.1% of fully supervised clDice. These results demonstrate the potential of foundation-model label propagation for highly label-efficient retinal vessel segmentation. Code is available at https://github.com/IAANNH/ESRVS.

arXiv

RP-OPSD: Resolution-Privileged On-Policy Self-Distillation for Multimodal Large Language Models

Qihui Zhu, Yuchen Wang, Zijian Wen, Tao Zhang, Mengjie Zhang, Yang Liu, Shuangwu Chen, Siying Wu, Jian Yang, Xiaofeng Jiang

On-Policy Self-Distillation (OPSD) uses privileged information available only to the teacher to provide dense token-level supervision on trajectories generated by the student. However, existing methods often rely on verified solution traces, explanations generated by external models, or manually localized visual evidence, which limits their scalable application to multimodal large language models. To address this issue, we exploit the information gap between high- and low-resolution views of the same image and propose RP-OPSD (Resolution-Privileged On-Policy Self-Distillation for Multimodal Large Language Models). During training, the student policy generates on-policy trajectories from images at one-quarter of the original resolution, while the teacher policy provides supervision using the original-resolution images. By minimizing the divergence between their output distributions along the student trajectories, the student learns the predictive behavior of the teacher under high-resolution inputs, thereby strengthening its low-resolution capability and transferring the learned improvement to original-resolution inference. RP-OPSD requires neither additional human annotations nor external models to generate solution traces but only image--question pairs. Experiments on Qwen3.5-9B show that RP-OPSD achieves a 5.45\% relative improvement in average performance at the original resolution and a $1.78\times$ training speedup over OPSD. These results demonstrate that resolution differences can serve as a simple and scalable source of privileged information, providing an effective and efficient approach to on-policy self-distillation for multimodal large language models.

arXiv

MSVS-VAE: Multi-Scale Anchored VecSet for High-Fidelity 3D Reconstruction

Dehao Hao, Kaiyi Zhang, Tanghui Jia, Xiangjun Gao, Dongyu Yan, Weikai Chen, Zeyu HU, Lingting Zhu, Yingda Yin, Runze Zhang, Li Yuan, Xin Wang, Long Quan

High-fidelity 3D generative modeling increasingly relies on the latent diffusion paradigm, where the reconstruction quality of the underlying 3D VAE becomes a primary bottleneck. Existing approaches largely follow two paradigms: sparse voxel-based representations achieve strong reconstruction quality but incur significant memory and computational overhead, while set-based representations are compact and continuous yet typically lag in fidelity due to latent sparsity and excessive global smoothness. We propose MSVS-VAE, a hierarchical set-based VAE that closes this fidelity gap without sacrificing compactness. Our key idea is to progressively densify anchored VecSet latents via hierarchical point-shuffle upsampling, increasing spatial capacity for fine-grained geometry modeling. To efficiently decode from the densified hierarchy, we replace global cross-attention with AVS-Conv, a geometry-aware local aggregation operator operating within local neighborhoods rather than the exhaustive latent set. We further introduce multi-scale query decoding to fuse coarse-to-fine latent features, where coarse scales provide stable global context, and fine scales refine localized geometry, reducing artifacts from overly local receptive fields. Extensive experiments on Objaverse, ABO, and in-the-wild benchmarks demonstrate that MSVS-VAE consistently outperforms prior set-based and voxel-based VAEs, delivering approximately 10x faster decoding than prior set-based methods and approximately 10x higher compactness than voxel-based baselines.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.