← 전체 목록

2026-11-07 AI 리서치 브리핑

최신 VLM, sLLM, on-device AI 논문과 연구 블로그를 한눈에 정리합니다. 중복 기사 방지를 위해 URL 기준으로 추적합니다.

총 10건 요약 자동 생성

VLM 업데이트

멀티모달 비전-언어 모델의 최신 논문과 리더보드 변화

Do You Miss the Nights We Spent Crafting Figures? Rethinking Teaser and Framework Figures in the Age of Generative AI

Paper arXiv cs.CV (recent)

Scientific figures give research ideas a visual form through choices of abstraction, composition, and emphasis. Generative AI offers new ways to realize these choices, raising questions about explanatory clarity and purposeful visual expression. We examine these questions through a visual retrospective of 371 motivation and framework figures from 200 papers. The corpus comprises a quota-based main sample of 160 papers spanning four temporal periods and a separate purposive style-contrast sample of 40 papers. A single multimodal judge evaluates six dimensions of communication and design, with three repeated judgments per figure. We attempt source-constrained revisions of all 371 figures, of which 369 pass the recorded fidelity checks. Original-figure ratings show no uniform decline across periods. Revisions increase mean clarity ratings by 0.428 points and decrease aesthetic coherence ratings by 0.111 points on a five-point scale. Gains in the six-dimension average are larger for main-sample motivation figures than for frameworks. Case analyses show that added explanation can clarify comparisons while leaving typographic hierarchy and visual economy unresolved. These model-based findings support researcher-led design, with AI assisting exploration and authors retaining responsibility for explanatory purpose, composition, and scientific relationships. The accompanying atlas and evaluation records are available at https://ntu-ai4x.github.io/Rethinking_Figures/index_en.html.

원문 보기

SAGE: A hybrid vision-language and geometry framework for work-zone pathway prediction under temporary traffic control

Paper arXiv cs.CV (recent)

이 논문은 임시 교통 통제 구역에서 자율주행 차량의 경로를 예측하기 위한 하이브리드 비전-언어 및 기하학 프레임워크인 SAGE를 제안합니다. SAGE는 비전-언어 모델의 궤적을 기하학적 경로 후보에 대한 선호도로 활용하며, MUTCD 규정을 기반으로 제어 준수 여부를 통합합니다. 실험 결과 SAGE는 기존 방법 대비 변위 오차와 제약 위반을 크게 줄이며, 안전성 및 정확도 측면에서 다른 주행 VLM보다 우수한 성능을 보였습니다.

English This paper proposes SAGE, a hybrid vision-language and geometry framework for predicting work-zone pathways for automated vehicles under temporary traffic control. SAGE utilizes a vision-language model's trajectory as a preference for geometric path candidates and integrates control compliance based on MUTCD provisions. Experiments demonstrate that SAGE significantly reduces displacement errors and hard-constraint violations, outperforming other driving VLMs in safety and accuracy.

원문 보기

On-Device AI

디바이스 내 추론 및 엣지 최적화 동향

EgoGeneralist: Unifying Egocentric Action-Video Tasks through Video-Native Modeling

Paper arXiv cs.CV (recent)

Modern video models possess strong visual priors, yet do not natively represent human action in metric 3D space. We present EgoGeneralist, which adapts pretrained video generators to unify action-conditioned egocentric video generation, pose estimation, and joint action-video generation. Our role-adaptive action renderer provides video-native representations for multidirectional action-video learning, making metric motion usable as both a spatial control signal and a recoverable generation target. Cross-stream action-video coupling establishes close correspondence between metric motion and visual dynamics through lightweight adaptation while preserving the pretrained visual prior. Together, these designs establish a generalist framework spanning five action-video tasks. Despite adaptation only on controlled laboratory recordings, EgoGeneralist generalizes to diverse unseen scenes and interactions with effective action control and visual quality, bridging pretrained visual knowledge and metric 3D human motion.

원문 보기

AI 뉴스 & 리서치

기업/연구기관의 주요 발표와 블로그 업데이트

VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding

Paper Hugging Face Papers

VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding에 관한 최근 업데이트입니다. 자세한 내용은 원문 링크에서 확인할 수 있습니다.

원문 보기

Harness-S: Learning from Sibling Harnesses for Generalizable Agent Training

Paper Hugging Face Papers

Harness-S: Learning from Sibling Harnesses for Generalizable Agent Training에 관한 최근 업데이트입니다. 자세한 내용은 원문 링크에서 확인할 수 있습니다.

원문 보기

Random Is Almost Enough? Stable and Low-Cost Benchmark Compression via BestRand

Paper Hugging Face Papers

Random Is Almost Enough? Stable and Low-Cost Benchmark Compression via BestRand에 관한 최근 업데이트입니다. 자세한 내용은 원문 링크에서 확인할 수 있습니다.

원문 보기

SLA: Scaling Spatial-Language-Action Model via Continual Pre-training

Paper Hugging Face Papers

SLA: Scaling Spatial-Language-Action Model via Continual Pre-training에 관한 최근 업데이트입니다. 자세한 내용은 원문 링크에서 확인할 수 있습니다.

원문 보기

WeLM-Image: A Unified Image Generation Model towards Visual Intelligence

Paper Hugging Face Papers

WeLM-Image: A Unified Image Generation Model towards Visual Intelligence에 관한 최근 업데이트입니다. 자세한 내용은 원문 링크에서 확인할 수 있습니다.

원문 보기

Where and Why is Grounded Reasoning Useful?

Paper arXiv cs.CV (recent)

Grounded reasoning, where a model outputs text coordinates of relevant locations in chain-of-thought reasoning, is a common strategy for solving complex visual reasoning problems. However, the benefits and limitations of grounded reasoning are understudied. We study where and why grounded reasoning is most useful. To do so, we introduce a chain-of-thought analysis method to measure how much a model uses its outputted grounding in its thinking trace and perform a detailed ablation study of a grounded reasoning training recipe. We find that grounded reasoning improves counting but shows no significant gain over baselines without coordinate grounding across a wide range of spatial reasoning and perception benchmarks. Using several synthetic tasks, we show that grounded reasoning is most useful when a task requires undescribable references, which are visual concepts that are difficult to describe in language. However, we observe that many existing visual reasoning benchmarks are solvable with only describable text references. To close this gap, we introduce BrickVision, a new benchmark that requires undescribable visual references to solve.

원문 보기

FluxFlow: Conservative Flow-Matching for Astronomical Image Super-Resolution

Paper arXiv cs.CV (recent)

이 논문은 지상 관측 데이터에서 우주 품질의 천문 이미지를 복원하는 초해상도 문제를 다룹니다. 저자들은 관측 불확실성과 소스 영역 중요도를 통합하고 환각 소스를 억제하는 FluxFlow라는 보수적인 픽셀 공간 플로우 매칭 프레임워크를 제안합니다. FluxFlow는 새로 구축된 대규모 DESI--HST 데이터셋을 사용하여 기존 방법보다 측광 및 과학적 정확도에서 일관되게 우수한 성능을 보였습니다.

English This paper addresses astronomical super-resolution, aiming to recover space-quality images from ground-based observations limited by resolution and atmospheric seeing. It introduces FluxFlow, a conservative pixel-space flow-matching framework that incorporates observation uncertainty and source-region importance weights, along with a training-free correction to suppress hallucination. Evaluated on the new DESI--HST Dataset, FluxFlow consistently outperforms existing baselines in both photometric and scientific accuracy.

원문 보기

참고한 소스