Reflect-R1 introduces the first evidence-driven self-correction framework for long video understanding using a three-stage pipeline, stage-decoupled RL via SD-GRPO, and a 120K dataset to achieve SOTA on VideoMME and LongVideoBench.
hub Baseline reference
VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning
Baseline reference. 50% of citing Pith papers use this work as a benchmark or comparison.
64
Pith papers citing it
Baseline
50% of classified citations
abstract
Recently, slow-thinking systems like GPT-o1 and DeepSeek-R1 have demonstrated great potential in solving challenging problems through explicit reflection. They significantly outperform the best fast-thinking models, such as GPT-4o, on various math and science benchmarks. However, their multimodal reasoning capabilities remain on par with fast-thinking models. For instance, GPT-o1's performance on benchmarks like MathVista, MathVerse, and MathVision is similar to fast-thinking models. In this paper, we aim to enhance the slow-thinking capabilities of vision-language models using reinforcement learning (without relying on distillation) to advance the state of the art. First, we adapt the GRPO algorithm with a novel technique called Selective Sample Replay (SSR) to address the vanishing advantages problem. While this approach yields strong performance, the resulting RL-trained models exhibit limited self-reflection or self-verification. To further encourage slow-thinking, we introduce Forced Rethinking, which appends a rethinking trigger token to the end of rollouts in RL training, explicitly enforcing a self-reflection reasoning step. By combining these two techniques, our model, VL-Rethinker, advances state-of-the-art scores on MathVista, MathVerse to achieve 80.4%, 63.5% respectively. VL-Rethinker also achieves open-source SoTA on multi-disciplinary benchmarks such as MathVision, MMMU-Pro, EMMA, and MEGA-Bench, narrowing the gap with OpenAI-o1. Our empirical results show the effectiveness of our approaches.
hub tools
citation-role summary
background 8
baseline 5
dataset 3
citation-polarity summary
representative citing papers
Introduces OCR-Robust benchmark and evaluates 18 VLMs showing clean accuracy does not guarantee robustness with charts and tables more fragile than documents under selected perturbations.
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite with largest gains at the 0.8B scale.
VA-OPD improves VLM performance over standard on-policy distillation by reweighting rollouts and separating KL terms according to token-level visual advantage on math and visual benchmarks.
Formalizes Reasoning Portability (RP) and proposes RDB-CL to modulate per-sample KL regularization in RLVR for MLLM continual learning, achieving +12.0% Last accuracy over vanilla RLVR baseline by preserving reusable reasoning on high-RP samples.
VLMs fail to detect image swaps during self-reflective reasoning with accuracy drops up to 60%, revealing that self-generated reflections do not trigger genuine visual re-examination.
GRASP is a large-scale dataset and benchmark for social reasoning grounded in gaze and gesture events in multi-person videos, with Social Grounding Reward (SGR) proposed to improve model performance on GRASP-Bench.
Visual debiasing of omni-modal benchmarks combined with staged post-training lets a 3B model match or exceed a 30B model without a stronger teacher.
Laser reformulates visual reasoning via Dynamic Windowed Alignment Learning to maintain latent superposition of global features, delivering 5.03% average gains over Monet and over 97% fewer inference tokens on six benchmarks.
GPRO trains a meta-controller on 790k failure-labeled samples to dynamically select fast, perception, or reasoning paths in LVLMs, yielding higher accuracy and shorter responses than prior slow-thinking methods.
DMLR performs dynamic visual-textual interleaving in latent space using confidence-guided latent policy gradient optimization and a dynamic visual injection strategy, yielding improved multimodal reasoning on benchmarks.
RS-EoT uses a SocraticAgent self-play system and two-stage RL to train VLMs for genuine iterative reasoning and visual inspection on remote sensing VQA and grounding tasks, achieving SOTA results.
Latent Visual Reasoning enables autoregressive generation of latent visual states that reconstruct critical image tokens, yielding gains on perception-heavy VQA benchmarks such as 71.67% on MMVP.
VGR introduces a visual-grounded reasoning MLLM that detects and replays image regions during inference, achieving gains on visual benchmarks with 30% fewer image tokens than the LLaVA-NeXT-7B baseline.
WebEye benchmark and Pixel-Searcher agent enable visual perception tasks by using web search to resolve object identities before precise localization or answering.
RAPO uses an information-theoretic lower bound on visual gain to select high-entropy reflection anchors and optimizes a chain-masked KL surrogate, delivering gains over baselines on reasoning benchmarks across LVLM backbones.
SeePhys Pro benchmark reveals multimodal models degrade on physics reasoning as information transfers from text to images, with blind training improvements often stemming from textual cues rather than visual evidence.
Perceval is a perception-centric PRM that detects token-level perceptual errors in VLMs, supporting token-advantage RL training and iterative test-time scaling for improved reasoning.
RL post-training on hallucination-forced multimodal data improves reasoning performance and can outperform standard training.
SegAnswer trains an MLLM to generate segmentation masks instead of bounding boxes when zooming into image regions during visual reasoning, yielding consistent improvements across perception and hallucination benchmarks.
Scene-graph-aligned SFT plus node-as-proxy GRPO rewards let small MLLMs outperform larger baselines on fine-grained visual reasoning tasks.
PRP introduces proactive routing via Draft Rating Learning and Joint Rating Learning to route queries early between draft and target models for efficient multimodal reasoning.
CFPO is a counterfactual policy optimization method that regularizes RL policies in LVLMs by maximizing prediction discrepancy under suppressed visual cues, reporting 3-6% gains over baselines.
VEPA adds an intermediate sufficiency-driven RL stage with GRPO to strengthen visual evidence grounding in MLLMs, yielding gains on visually demanding benchmarks that transfer beyond task-specific training.
citing papers explorer
-
Reflect-R1: Evidence-Driven Reflection for Self-Correction in Long Video Understanding
Reflect-R1 introduces the first evidence-driven self-correction framework for long video understanding using a three-stage pipeline, stage-decoupled RL via SD-GRPO, and a 120K dataset to achieve SOTA on VideoMME and LongVideoBench.
-
How Robust is OCR-Reasoning? Evaluating OCR-Reasoning Robustness of Vision-Language Models under Visual Perturbations
Introduces OCR-Robust benchmark and evaluates 18 VLMs showing clean accuracy does not guarantee robustness with charts and tables more fragile than documents under selected perturbations.
-
Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite with largest gains at the 0.8B scale.
-
Visual-Advantage On-Policy Distillation for Vision-Language Models
VA-OPD improves VLM performance over standard on-policy distillation by reweighting rollouts and separating KL terms according to token-level visual advantage on math and visual benchmarks.
-
Reasoning Portability: Guiding Continual Learning for MLLMs in the RLVR Era
Formalizes Reasoning Portability (RP) and proposes RDB-CL to modulate per-sample KL regularization in RLVR for MLLM continual learning, achieving +12.0% Last accuracy over vanilla RLVR baseline by preserving reusable reasoning on high-RP samples.
-
Are VLMs Seeing or Just Saying? Uncovering the Illusion of Visual Re-examination
VLMs fail to detect image swaps during self-reflective reasoning with accuracy drops up to 60%, revealing that self-generated reflections do not trigger genuine visual re-examination.
-
GRASP: Learning to Ground Social Reasoning in Multi-Person Non-Verbal Interactions
GRASP is a large-scale dataset and benchmark for social reasoning grounded in gaze and gesture events in multi-person videos, with Social Grounding Reward (SGR) proposed to improve model performance on GRASP-Bench.
-
Boosting Omni-Modal Language Models: Staged Post-Training with Visually Debiased Evaluation
Visual debiasing of omni-modal benchmarks combined with staged post-training lets a 3B model match or exceed a 30B model without a stronger teacher.
-
Forest Before Trees: Latent Superposition for Efficient Visual Reasoning
Laser reformulates visual reasoning via Dynamic Windowed Alignment Learning to maintain latent superposition of global features, delivering 5.03% average gains over Monet and over 97% fewer inference tokens on six benchmarks.
-
Addressing Overthinking in Large Vision-Language Models via Gated Perception-Reasoning Optimization
GPRO trains a meta-controller on 790k failure-labeled samples to dynamically select fast, perception, or reasoning paths in LVLMs, yielding higher accuracy and shorter responses than prior slow-thinking methods.
-
Reasoning Within the Mind: Dynamic Multimodal Interleaving in Latent Space
DMLR performs dynamic visual-textual interleaving in latent space using confidence-guided latent policy gradient optimization and a dynamic visual injection strategy, yielding improved multimodal reasoning on benchmarks.
-
Asking like Socrates: Socrates helps VLMs understand remote sensing images
RS-EoT uses a SocraticAgent self-play system and two-stage RL to train VLMs for genuine iterative reasoning and visual inspection on remote sensing VQA and grounding tasks, achieving SOTA results.
-
Latent Visual Reasoning
Latent Visual Reasoning enables autoregressive generation of latent visual states that reconstruct critical image tokens, yielding gains on perception-heavy VQA benchmarks such as 71.67% on MMVP.
-
VGR: Visual Grounded Reasoning
VGR introduces a visual-grounded reasoning MLLM that detects and replays image regions during inference, achieving gains on visual benchmarks with 30% fewer image tokens than the LLaVA-NeXT-7B baseline.
-
From Web to Pixels: Bringing Agentic Search into Visual Perception
WebEye benchmark and Pixel-Searcher agent enable visual perception tasks by using web search to resolve object identities before precise localization or answering.
-
Reflection Anchors for Propagation-Aware Visual Retention in Long-Chain Multimodal Reasoning
RAPO uses an information-theoretic lower bound on visual gain to select high-entropy reflection anchors and optimizes a chain-masked KL surrogate, delivering gains over baselines on reasoning benchmarks across LVLM backbones.
-
SeePhys Pro: Diagnosing Modality Transfer and Blind-Training Effects in Multimodal RLVR for Physics Reasoning
SeePhys Pro benchmark reveals multimodal models degrade on physics reasoning as information transfers from text to images, with blind training improvements often stemming from textual cues rather than visual evidence.
-
Improving Vision-language Models with Perception-centric Process Reward Models
Perceval is a perception-centric PRM that detects token-level perceptual errors in VLMs, supporting token-advantage RL training and iterative test-time scaling for improved reasoning.
-
Understanding the Role of Hallucination in Reinforcement Post-Training of Multimodal Reasoning Models
RL post-training on hallucination-forced multimodal data improves reasoning performance and can outperform standard training.
-
Segmentation before Answering: Pixel Grounding for MLLM Visual Reasoning
SegAnswer trains an MLLM to generate segmentation masks instead of bounding boxes when zooming into image regions during visual reasoning, yielding consistent improvements across perception and hallucination benchmarks.
-
Scene Graph Thinking: Reinforcing Structured Visual Reasoning for Multimodal Large Language Models
Scene-graph-aligned SFT plus node-as-proxy GRPO rewards let small MLLMs outperform larger baselines on fine-grained visual reasoning tasks.
-
Before Thinking, Learn to Decide: Proactive Routing for Efficient Visual Reasoning
PRP introduces proactive routing via Draft Rating Learning and Joint Rating Learning to route queries early between draft and target models for efficient multimodal reasoning.
-
CFPO: Counterfactual Policy Optimization for Multimodal Reasoning
CFPO is a counterfactual policy optimization method that regularizes RL policies in LVLMs by maximizing prediction discrepancy under suppressed visual cues, reporting 3-6% gains over baselines.
-
See First, Answer Later: Visual Evidence Pre-Alignment via Sufficiency-Driven RL
VEPA adds an intermediate sufficiency-driven RL stage with GRPO to strengthen visual evidence grounding in MLLMs, yielding gains on visually demanding benchmarks that transfer beyond task-specific training.
-
Reinforcing Dual-Path Reasoning in Spatial Vision Language Models
SR-REAL equips spatial VLMs with dual LOR and DTR reasoning paths trained via RL, achieving better benchmark performance through mutual reinforcement and generalization without per-task tuning.
-
Fine-grained Fragment Retrieval in Multi-modal Long-form Dialogues
Introduces FFR task, F2RVLM and FFRS models, and MLDR dataset for retrieving coherent multi-modal dialogue fragments, reporting superior performance on single-dialogue and corpus benchmarks.
-
TRON: Targeted Rule-Verifiable Online Environments for Visual Reasoning RL
TRON supplies 520 rule-verifiable online visual reasoning environments across five ability buckets that generate unlimited training instances for RL post-training, yielding consistent gains on ten external multimodal benchmarks for three vision-language models.
-
Zamba2-VL Technical Report
Zamba2-VL is a family of 1.2B–7B hybrid Mamba2-transformer vision-language models that match leading transformer VLMs on image, reasoning, OCR, grounding and counting benchmarks while delivering roughly 10x lower time-to-first-token.
-
Guidance Contrastive Token Credit Assignment for Discrete Policy Optimization
GCPO performs per-token credit assignment in discrete policy optimization by setting token advantages proportional to the difference in model predictions under positive versus negative prompts, outperforming GRPO and DAPO on text-to-image and chain-of-thought tasks.
-
Agent Explorative Policy Optimization for Multimodal Agentic Reasoning
AXPO addresses the Thinking-Acting Gap in agentic RL training by targeted resampling of tool calls in all-wrong subgroups, delivering +1.8pp gains over GRPO on nine multimodal benchmarks with an 8B model beating a 32B baseline on Pass@4.
-
AnE: Pushing the Reasoning Frontier of Multimodal LLMs via Anchor Evolution
AnE combines Truth Anchor Expansion and Scaffold-Stripping to deliver 10.3% gains on eight multimodal reasoning benchmarks for MLLMs.
-
Learning Spatiotemporal Sensitivity in Video LLMs via Counterfactual Reinforcement Learning
CRPO applies counterfactual videos and a cross-branch relation reward in RL post-training to reduce shortcut reliance in Video LLMs, with gains shown on the new DyBench paired benchmark.
-
Leveraging Latent Visual Reasoning in Silence
Latent visual reasoning improves multimodal models via training effects even without using latent tokens at inference, enabled by an attention-based RL reward that promotes interaction with text tokens.
-
PanoWorld: Towards Spatial Supersensing in 360$^\circ$ Panorama World
PanoWorld adds spherical spatial cross-attention and pano-native training data to MLLMs for improved spatial reasoning on ERP panoramas, outperforming baselines on new and existing benchmarks.
-
G$^2$TR: Generation-Guided Visual Token Reduction for Separate-Encoder Unified Multimodal Models
G²TR reduces visual tokens and prefill compute by 1.94x in separate-encoder UMMs via generation-guided importance from VAE latent consistency, balanced selection, and merging, while preserving reasoning accuracy and editing quality.
-
HyLaR: Hybrid Latent Reasoning with Decoupled Policy Optimization
HyLaR interleaves discrete text generation with continuous visual latent representations and optimizes them via a decoupled RL algorithm using vMF distributions, improving fine-grained visual reasoning.
-
Boosting Reasoning in Large Multimodal Models via Activation Replay
Activation Replay boosts multimodal reasoning in post-trained LMMs by replaying low-entropy activations from base models to RLVR counterparts at test time via visual token manipulation.
-
REVISOR: Beyond Textual Reflection, Towards Multimodal Introspective Reasoning in Long-Form Video Understanding
REVISOR adds multimodal visual-text reflection and a Dual Attribution Decoupled Reward to improve long-form video reasoning in MLLMs without extra supervised fine-tuning.
-
DeepEyesV2: Toward Agentic Multimodal Model
DeepEyesV2 uses a two-stage cold-start plus reinforcement learning pipeline to produce an agentic multimodal model that adaptively invokes tools and outperforms direct RL on real-world reasoning benchmarks.
-
Does RLVR Extend Reasoning Boundaries? Investigating Capability Expansion in Vision-Language Models
RLVR on synthetic mazes enables VLMs to solve spatial reasoning tasks unreachable by the base model and generalizes to real-world navigation benchmarks.
-
Locate-Then-Examine: Grounded Region Reasoning Improves Detection of AI-Generated Images
Locate-Then-Examine improves AI-generated image detection by localizing suspicious regions first then performing region-aware re-examination, while releasing the TRACE dataset of 20k annotated images.
-
Embodied-R1: Reinforced Embodied Reasoning for General Robotic Manipulation
Embodied-R1 uses a pointing-centric representation and reinforced fine-tuning on a 200K dataset to achieve state-of-the-art results on embodied benchmarks plus 56.2% success in SIMPLEREnv and 87.5% on real XArm tasks without task-specific training.
-
Perception-Aware Policy Optimization for Multimodal Reasoning
PAPO integrates perception-aware supervision via a KL-based loss into RLVR methods like GRPO, yielding 4.4-17.5% gains on multimodal benchmarks and 30.5% fewer perception errors, with larger gains on vision-heavy tasks.
-
OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles
Iterative SFT-RL cycles enable a 7B LVLM to develop sophisticated visual chain-of-thought reasoning and improve performance on math and general reasoning benchmarks.
-
Seir\^enes: Adversarial Self-Play with Evolving Distractions for LLM Reasoning
Seirênes trains LLMs via adversarial self-play to generate and overcome evolving distractions, producing gains of 7-10 points on math reasoning benchmarks and exposing blind spots in larger models.
-
Reinforcing Multimodal Reasoning Against Visual Degradation
ROMA improves MLLM robustness to seen and unseen visual corruptions by +2.3-2.4% over GRPO on seven reasoning benchmarks while matching clean accuracy.
-
Persistent Visual Memory: Sustaining Perception for Deep Generation in LVLMs
PVM adds a parallel branch to LVLMs that directly supplies visual embeddings to prevent attention decay over long generated sequences, yielding accuracy gains on reasoning tasks with minimal overhead.
-
DR-MMSearchAgent: Deepening Reasoning in Multimodal Search Agents
DR-MMSearchAgent derives batch-wide trajectory advantages and uses differentiated Gaussian rewards to prevent premature collapse in multimodal agents, outperforming MMSearch-R1 by 8.4% on FVQA-test.
-
SceneCritic: A Symbolic Evaluator for 3D Indoor Scene Synthesis
SceneCritic is a symbolic, ontology-grounded evaluator for floor-plan layouts that identifies specific semantic, orientation, and geometric violations and aligns better with human judgments than VLM-based scorers.
-
Visual Enhanced Depth Scaling for Multimodal Latent Reasoning
Visual replay module and adaptive depth scaling improve multimodal latent reasoning, reaching SOTA benchmarks with faster inference than explicit chain-of-thought methods.