FigSIM is the first annotated dataset for fine-grained suicide severity and figurative language in suicide memes, accompanied by benchmarks on 16 unimodal and multimodal models.
super hub Mixed citations
InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency
Mixed citation behavior. Most common role is background (50%).
440
Pith papers citing it
3
external citations · Pith
Background
50% of classified citations
abstract
We introduce InternVL 3.5, a new family of open-source multimodal models that significantly advances versatility, reasoning capability, and inference efficiency along the InternVL series. A key innovation is the Cascade Reinforcement Learning (Cascade RL) framework, which enhances reasoning through a two-stage process: offline RL for stable convergence and online RL for refined alignment. This coarse-to-fine training strategy leads to substantial improvements on downstream reasoning tasks, e.g., MMMU and MathVista. To optimize efficiency, we propose a Visual Resolution Router (ViR) that dynamically adjusts the resolution of visual tokens without compromising performance. Coupled with ViR, our Decoupled Vision-Language Deployment (DvD) strategy separates the vision encoder and language model across different GPUs, effectively balancing computational load. These contributions collectively enable InternVL3.5 to achieve up to a +16.0\% gain in overall reasoning performance and a 4.05$\times$ inference speedup compared to its predecessor, i.e., InternVL3. In addition, InternVL3.5 supports novel capabilities such as GUI interaction and embodied agency. Notably, our largest model, i.e., InternVL3.5-241B-A28B, attains state-of-the-art results among open-source MLLMs across general multimodal, reasoning, text, and agentic tasks -- narrowing the performance gap with leading commercial models like GPT-5. All models and code are publicly released.
hub tools
citation-role summary
background 55
baseline 41
method 12
dataset 1
other 1
citation-polarity summary
claims ledger
- abstract We introduce InternVL 3.5, a new family of open-source multimodal models that significantly advances versatility, reasoning capability, and inference efficiency along the InternVL series. A key innovation is the Cascade Reinforcement Learning (Cascade RL) framework, which enhances reasoning through a two-stage process: offline RL for stable convergence and online RL for refined alignment. This coarse-to-fine training strategy leads to substantial improvements on downstream reasoning tasks, e.g., MMMU and MathVista. To optimize efficiency, we propose a Visual Resolution Router (ViR) that dynami
authors
co-cited works
representative citing papers
EPIC-Bench is a new fine-grained benchmark that shows leading VLMs struggle with multi-target counting, part-whole relations, and affordance detection in real-world embodied visual grounding tasks.
LongAct benchmark evaluates long-horizon household task execution from free-form instructions; HoloMind agent raises performance but top VLMs still reach only 59% goal completion and 16% full-task success.
SenseBench is the first physics-based benchmark with 10K+ instances and dual protocols to evaluate VLMs on remote sensing low-level perception and diagnostic description, revealing domain bias and specific failure modes.
EgoMemReason is a new benchmark showing that even the best multimodal models achieve only 39.6% accuracy on reasoning tasks that require integrating sparse evidence across days in egocentric video.
RuleSafe-VL creates 2,166 rule-conditioned cases from 93 atomic rules and 92 relations across three policy families to diagnose where VLMs fail at rule-based content moderation reasoning.
TraceAV-Bench is the first benchmark for multi-hop trajectory reasoning over long audio-visual videos, showing top models reach only 51-68% accuracy with substantial room for improvement.
MedHorizon benchmark reveals current multimodal LLMs achieve only 41.1% accuracy on long medical videos due to failures in sparse evidence retrieval and procedural reasoning.
Creates the first benchmark dataset integrating papers, slides, videos, and presentations for evaluating AI models on fine-grained multimodal correspondences in science.
MLLMs exhibit a Mirage effect by bypassing circuit diagrams in favor of header semantics for Verilog generation; VeriGround with identifier anonymization and D-ORPO training reaches 46% Functional Pass@1 while refusing blank images at >92%.
MM-JudgeBench shows substantial cross-lingual performance variance in 22 LVLM judges, with model size and architecture as poor predictors of multilingual robustness.
VLMs hallucinate by prioritizing contradictory on-screen text over visual content, addressed via the VisualTextTrap benchmark with 6,057 human-validated samples and the VTHM-MoE dual-encoder framework using dimension-specific experts and adaptive routing.
RefereeBench shows that even the strongest video MLLMs reach only around 60% accuracy on multi-sport refereeing tasks and struggle with rule application and temporal grounding.
Dental-TriageBench is the first expert-annotated multimodal benchmark for hierarchical dental triage and shows a substantial performance gap between 19 MLLMs and junior dentists, especially on multi-domain referral cases.
VAREX benchmark shows structured output compliance limits models under 4B parameters more than extraction ability, with layout-preserving text giving the largest accuracy gains over images.
Molmo2 delivers state-of-the-art open-weight video VLMs with new grounding datasets and training methods that outperform prior open models and match or exceed some proprietary ones on pointing and tracking tasks.
ToG-Bench is the first benchmark for task-oriented spatio-temporal video grounding in egocentric videos, with explicit-implicit dual grounding and one-to-many object scenarios across 100 ScanNet clips and 2704 instructions.
MTLA is a training-free, post-hoc confidence score for multimodal LLM localization that restricts attention aggregation to the model's own predicted region and tokens, substantially improving hallucination detection and re-ranking across image, video, and audio.
AnyGroundBench is a domain-adaptation benchmark for spatio-temporal video grounding across animal, industry, sports, surgery, and public security domains that finds 15 state-of-the-art VLMs fail in zero-shot and ICL settings.
LongEgoRefer is a new benchmark of 1,498 referring expressions in 45-minute average egocentric videos that exposes the failure of existing Video REC models on sparse long-form spatio-temporal grounding.
A large examination-level ultrasound dataset with long-form reports enables simple LVLM fine-tuning to outperform prior complex methods.
Introduces Anom-π framework for active video anomaly understanding via interleaved policy optimization and iDPO under weak supervision, claiming a 2B model outperforms larger SOTA VAU models.
EgoGapBench shows humans reliably select egocentric actions in multi-agent scenes while MLLMs systematically choose other agents' actions, and standard egocentric training data fails to close the gap.
EgoSafetyBench shows VLMs reliably spot hazard-containing videos but miss specific contextual hazards and are degraded by misleading in-scene text.
citing papers explorer
-
FigSIM: A Dataset for Fine-grained Suicide Severity and Figurative Language in Suicide Memes
FigSIM is the first annotated dataset for fine-grained suicide severity and figurative language in suicide memes, accompanied by benchmarks on 16 unimodal and multimodal models.
-
EPIC-Bench: A Perception-Centric Benchmark for Fine-Grained Embodied Visual Grounding in Vision-Language Models
EPIC-Bench is a new fine-grained benchmark that shows leading VLMs struggle with multi-target counting, part-whole relations, and affordance detection in real-world embodied visual grounding tasks.
-
When Robots Do the Chores: A Benchmark and Agent for Long-Horizon Household Task Execution
LongAct benchmark evaluates long-horizon household task execution from free-form instructions; HoloMind agent raises performance but top VLMs still reach only 59% goal completion and 16% full-task success.
-
SenseBench: A Benchmark for Remote Sensing Low-Level Visual Perception and Description in Large Vision-Language Models
SenseBench is the first physics-based benchmark with 10K+ instances and dual protocols to evaluate VLMs on remote sensing low-level perception and diagnostic description, revealing domain bias and specific failure modes.
-
EgoMemReason: A Memory-Driven Reasoning Benchmark for Long-Horizon Egocentric Video Understanding
EgoMemReason is a new benchmark showing that even the best multimodal models achieve only 39.6% accuracy on reasoning tasks that require integrating sparse evidence across days in egocentric video.
-
RuleSafe-VL: Evaluating Rule-Conditioned Decision Reasoning in Vision-Language Content Moderation
RuleSafe-VL creates 2,166 rule-conditioned cases from 93 atomic rules and 92 relations across three policy families to diagnose where VLMs fail at rule-based content moderation reasoning.
-
TraceAV-Bench: Benchmarking Multi-Hop Trajectory Reasoning over Long Audio-Visual Videos
TraceAV-Bench is the first benchmark for multi-hop trajectory reasoning over long audio-visual videos, showing top models reach only 51-68% accuracy with substantial room for improvement.
-
MedHorizon: Towards Long-context Medical Video Understanding in the Wild
MedHorizon benchmark reveals current multimodal LLMs achieve only 41.1% accuracy on long medical videos due to failures in sparse evidence retrieval and procedural reasoning.
-
Unifying Scientific Communication: Fine-Grained Correspondence Across Scientific Media
Creates the first benchmark dataset integrating papers, slides, videos, and presentations for evaluating AI models on fine-grained multimodal correspondences in science.
-
From Mirage to Grounding: Towards Reliable Multimodal Circuit-to-Verilog Code Generation
MLLMs exhibit a Mirage effect by bypassing circuit diagrams in favor of header semantics for Verilog generation; VeriGround with identifier anonymization and D-ORPO training reaches 46% Functional Pass@1 while refusing blank images at >92%.
-
Lost in Translation: Do LVLM Judges Generalize Across Languages?
MM-JudgeBench shows substantial cross-lingual performance variance in 22 LVLM judges, with model size and architecture as poor predictors of multilingual robustness.
-
When Text Hijacks Vision: Benchmarking and Mitigating Text Overlay-Induced Hallucination in Vision Language Models
VLMs hallucinate by prioritizing contradictory on-screen text over visual content, addressed via the VisualTextTrap benchmark with 6,057 human-validated samples and the VTHM-MoE dual-encoder framework using dimension-specific experts and adaptive routing.
-
RefereeBench: Are Video MLLMs Ready to be Multi-Sport Referees
RefereeBench shows that even the strongest video MLLMs reach only around 60% accuracy on multi-sport refereeing tasks and struggle with rule application and temporal grounding.
-
Dental-TriageBench: Benchmarking Multimodal Reasoning for Hierarchical Dental Triage
Dental-TriageBench is the first expert-annotated multimodal benchmark for hierarchical dental triage and shows a substantial performance gap between 19 MLLMs and junior dentists, especially on multi-domain referral cases.
-
VAREX: A Benchmark for Multi-Modal Structured Extraction from Documents
VAREX benchmark shows structured output compliance limits models under 4B parameters more than extraction ability, with layout-preserving text giving the largest accuracy gains over images.
-
Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding
Molmo2 delivers state-of-the-art open-weight video VLMs with new grounding datasets and training methods that outperform prior open models and match or exceed some proprietary ones on pointing and tracking tasks.
-
ToG-Bench: Task-Oriented Spatio-Temporal Grounding in Egocentric Videos
ToG-Bench is the first benchmark for task-oriented spatio-temporal video grounding in egocentric videos, with explicit-implicit dual grounding and one-to-many object scenarios across 100 ScanNet clips and 2704 instructions.
-
Propose and Attend: Training-free MLLM Grounding Confidence via Multi-Token Localized Attention
MTLA is a training-free, post-hoc confidence score for multimodal LLM localization that restricts attention aggregation to the model's own predicted region and tokens, substantially improving hallucination detection and re-ranking across image, video, and audio.
-
AnyGroundBench: A Specialized-Domain Benchmark for Video Grounding in Vision-Language Models
AnyGroundBench is a domain-adaptation benchmark for spatio-temporal video grounding across animal, industry, sports, surgery, and public security domains that finds 15 state-of-the-art VLMs fail in zero-shot and ICL settings.
-
LongEgoRefer: A Benchmark for Long-Form Egocentric Video Referring Expression Comprehension
LongEgoRefer is a new benchmark of 1,498 referring expressions in 45-minute average egocentric videos that exposes the failure of existing Video REC models on sparse long-form spatio-temporal grounding.
-
Towards Real-World Ultrasound Understanding: Large Vision-Language Models from Multi-Image Examinations with Long-Form Reports
A large examination-level ultrasound dataset with long-form reports enables simple LVLM fine-tuning to outperform prior complex methods.
-
Learning to Watch: Active Video Anomaly Understanding via Interleaved Policy Optimization
Introduces Anom-π framework for active video anomaly understanding via interleaved policy optimization and iDPO under weak supervision, claiming a 2B model outperforms larger SOTA VAU models.
-
EgoGapBench: Benchmarking Egocentric Action Selection in Multi-Agent Scenes
EgoGapBench shows humans reliably select egocentric actions in multi-agent scenes while MLLMs systematically choose other agents' actions, and standard egocentric training data fails to close the gap.
-
EgoSafetyBench: A Diagnostic Egocentric Video Benchmark for Evaluating Embodied VLMs as Runtime Safety Guards
EgoSafetyBench shows VLMs reliably spot hazard-containing videos but miss specific contextual hazards and are degraded by misleading in-scene text.
-
Steal the Patch Size: Adversarially Manipulate Vision-Language Models
A side-channel attack infers ViT patch size from periodic accuracy collapses on aligned grid images, enabling preprocessing-aware transfer attacks on VLMs.
-
Visual Semantic Entropy: Do Vision Language Models Recognize Visual Ambiguity?
VSE perturbs images only to probe visual ambiguity in VLMs, clusters outputs into semantic prototypes, and computes mass-weighted dispersion, outperforming prior entropy methods on five VQA benchmarks across five models.
-
OmniCoT: A Benchmark for Global and Multi-Step Panoramic Reasoning
OmniCoT is a new panoramic reasoning benchmark with 6.7K eval, 1K real, and 14.3K training examples plus a two-stage SFT+GRPO training method to enforce global 360-degree consistency.
-
On Test-Time Scaling for Vision-Language Models
Small well-performing LVLMs gain the largest benefits from test-time scaling (up to ~30% improvement), often matching or exceeding larger models, while visual tokens contribute mainly early in the reasoning chain.
-
HumanMoveVQA: Can Video MLLMs reason about human movement in videos?
HumanMoveVQA is a new benchmark that generates 10K+ QA pairs from 3D-lifted video tracks to evaluate video MLLMs on global human trajectory and orientation reasoning.
-
SpatialUAV: Benchmarking Spatial Intelligence for Low-Altitude UAV Perception, Collaboration, and Motion
SpatialUAV releases a new multi-task benchmark for low-altitude UAV spatial intelligence and demonstrates that existing VLMs exhibit clear weaknesses in cross-view association and geometric reasoning.
-
Video-MME-Logical: A Controlled Diagnostic Benchmark for Video Temporal-Logical Reasoning
Introduces Video-MME-Logical benchmark for controlled diagnostic evaluation of temporal-logical reasoning in MLLMs via five operations and 25 fine-grained tasks.
-
Position Rebinding Cache Reuse: Replay-Free Visual Revisiting for Interleaved Multimodal Reasoning
PRCR enables replay-free visual revisiting in interleaved multimodal reasoning by storing raw visual KV caches with spatial coordinates and rebinding keys to position-compatible coordinates, matching replay performance while cutting computation by orders of magnitude.
-
DiCoBench: Benchmarking Multi-Image Fine-Grained Perception via Differential and Commonality Visual Cues
DiCoBench is a new high-resolution multi-image benchmark exposing large gaps between top MLLMs and human performance (98.3%) on differential and commonality visual cue perception.
-
Same Evidence, Different Answer: Auditing Order Sensitivity in Multimodal Large Language Models
All 18 audited MLLMs exhibit order sensitivity with per-facet flip rates of 24-50%, exceeding same-order decoder noise.
-
TriViewBench: Controlled Complexity Scaling for Multi-View Structural Reasoning in MLLMs
TriViewBench shows all tested MLLMs follow the same capability order with sharp drops on complex multi-view tasks and near-zero gain from Chain-of-Thought prompting.
-
SSMNBench: Diagnosing Image-based Cross-View Human-Object Understanding via Single-View Sufficiency and Multi-View Necessity
SSMNBench shows that MLLMs suffer distraction degradation on single-view-sufficient tasks and fail to integrate geometric evidence across views, instead relying on semantic averaging and view preference.
-
C3-Bench: A Context-Aware Change Captioning Benchmark
C3-Bench supplies a multi-domain dataset and LLM-based evaluation protocol that exposes systematic failures in existing change captioning models outside their training regimes.
-
Advancing WordArt-Oriented Scene Text Recognition: Datasets and Methods
Introduces 2M synthetic WATER-S dataset and WATERec model achieving 90.40% accuracy on WordArt-Bench, outperforming prior STR methods and VLMs.
-
video-SALMONN-R$^3$: Learning to ReWatch, ReAsk, and ReAnswer for Efficient Video Understanding
video-SALMONN-R³ is an end-to-end video-LLM trained with reinforcement learning to perform selective re-watching, re-asking, and re-answering for efficient video question answering without chain-of-thought cold-start supervision.
-
Mind the Heads: Topological Representation Alignment for Multimodal LLMs
HeRA aligns least-aligned attention heads in MLLMs using an MKNN-based contrastive objective to preserve cross-modal topological structure, yielding gains on vision-centric tasks and reduced hallucinations across 18 benchmarks.
-
RoboGaze: Evaluating Robot World Models via Structured Vision-Language Analysis
RoboGaze presents a structured multi-agent VLM pipeline and robotics-specific error taxonomy that improves video evaluation metrics by up to 43 F1 points over zero-shot baselines on a 382-clip dataset.
-
CheXpercept: A Benchmark for Evaluating Expert-Level Lesion Perception in Chest X-rays
CheXpercept is a sequential multi-level perception benchmark showing VLMs perform adequately only on coarse lesion detection in chest X-rays while degrading sharply on finer tasks, with medical VLMs offering no advantage over general models.
-
NEST: Narrative Event Structures in Time for Long Video Understanding
NEST is a new benchmark dataset for narrative event structures in long videos, with baselines reporting ETD below 8%, EL under 6%, EAE below 11%, and ERE at 35-44% F1.
-
PorTEXTO: A European Portuguese Benchmark for Visual Text Extraction
PorTEXTO benchmark shows sharp real-world performance drops in pt-PT OCR and finds specialized multilingual data outperforms model size or resolution increases.
-
MODE: Modality-Decomposed Expert-Level Mixed-Precision Quantization for MoE Multimodal LLMs
MODE decomposes expert selection frequency by modality, filters redundant vision tokens, adds per-modality sensitivity, and uses ILP to assign bit-widths, limiting average loss to 2.9% at W3A16 on MoE-MLLMs.
-
Natural-Language Temporal Grounding in Hour-Long Videos is a Search Problem: A Benchmark and Empirical Decomposition
Hour-long video temporal grounding is a search problem, shown by a new benchmark where all Video-LLMs collapse, frame retrieval outperforms them, 85% of failures are search-related, and a retrieve-then-ground hybrid improves results 6.7x.
-
3D-CoS: A New 3D Reconstruction Paradigm Based on VLM Code Synthesis
3D-CoS represents 3D objects as Blender code generated by VLMs, with workflows for planning, RAG, and agents, showing better edit fidelity than point-cloud baselines.
-
From Senses to Decisions: The Information Flow of Auditory and Visual Perception in Multimodal LLMs
AVLLMs route audio-visual information sequentially in video tasks and via parallel streams for interleaved items, allowing early token discard with little performance loss across models and scales.
-
When No Answer Is Correct: Diagnosing Absent Answer Detection for MLLMs in Video Understanding
MLLMs fail to detect absent correct answers in video QA tasks across three evaluation settings, defaulting to distractors even with chain-of-thought prompting.
-
Sci-Rho: A Multilingual Visually-Grounded Symbolic Benchmark for STEM Problems
Sci-Rho is a dynamic multilingual visually-grounded symbolic benchmark for STEM problems that reveals robustness gaps in current VLMs between average and worst-case performance.