P2R decouples perception from reasoning in VLMs via a two-stage process and PRA-GRPO alternating RL training, reporting gains such as 93.2% on V-Star for the 4B model over its Qwen3-VL backbone.
hub Canonical reference
GRIT: Teaching MLLMs to Think with Images
Canonical reference. 100% of citing Pith papers cite this work as background.
27
Pith papers citing it
Background
100% of classified citations
abstract
Recent studies have demonstrated the efficacy of using Reinforcement Learning (RL) in building reasoning models that articulate chains of thoughts prior to producing final answers. However, despite ongoing advances that aim at enabling reasoning for vision-language tasks, existing open-source visual reasoning models typically generate reasoning content with pure natural language, lacking explicit integration of visual information. This limits their ability to produce clearly articulated and visually grounded reasoning chains. To this end, we propose Grounded Reasoning with Images and Texts (GRIT), a novel method for training MLLMs to think with images. GRIT introduces a grounded reasoning paradigm, in which models generate reasoning chains that interleave natural language and explicit bounding box coordinates. These coordinates point to regions of the input image that the model consults during its reasoning process. Additionally, GRIT is equipped with a reinforcement learning approach, GRPO-GR, built upon the GRPO algorithm. GRPO-GR employs robust rewards focused on the final answer accuracy and format of the grounded reasoning output, which eliminates the need for data with reasoning chain annotations or explicit bounding box labels. As a result, GRIT achieves exceptional data efficiency, requiring as few as 20 image-question-answer triplets from existing datasets. Comprehensive evaluations demonstrate that GRIT effectively trains MLLMs to produce coherent and visually grounded reasoning chains, showing a successful unification of reasoning and grounding abilities.
hub tools
citation-role summary
background 5
citation-polarity summary
roles
background 5polarities
background 5representative citing papers
Proposes Spatial Narrative Score (SNS) evaluation for VLMs' camera motion understanding and introduces CaMo model achieving consistent performance on SNS and direct QA.
Mind's Eye benchmark shows top multimodal LLMs score below 50% on visual abstraction, relation, and transformation tasks while humans reach 80%.
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.
Scene-graph-aligned SFT plus node-as-proxy GRPO rewards let small MLLMs outperform larger baselines on fine-grained visual reasoning tasks.
Lens purifies visual evidence in MLLMs via question-conditioned latent noise masking with a LET token, yielding 2.4-6.4 point gains on VQA and grounding tasks.
SeProD is a plug-and-play self-prophetic decoding framework that combines pre- and post-training LVLM capabilities via probability-based sampling to improve coherent visual search and multi-step reasoning.
Introduces Zoom-then-Diagnose paradigm and uncertainty-aware reward in GRPO for confidence-aware ultrasound VQA, reporting 39.3% improvement in lesion localization across liver, breast, and thyroid datasets.
Perception Programs rewrite dense visual tool outputs into language-native summaries, boosting MLLM accuracy by 15-45% absolute on BLINK perception tasks and setting new state-of-the-art results.
TTSP samples and filters multiple zoom-in exploration traces and iteratively consolidates validated observations into an Evidence Ledger, improving fine-grained multimodal reasoning on V* Bench, HR-Bench, TreeBench, and MME-RealWorld-Lite.
SciTikZer-8B uses a new dataset, benchmark, and dual self-consistency RL to generate TikZ code for scientific graphics, outperforming much larger models like Gemini-2.5-Pro.
An RL-based questioner agent adaptively generates queries to discover novel failure modes in VLMs without human intervention.
CharTool equips MLLMs with cropping and code tools plus agentic RL on DuoChart data to raise chart-reasoning accuracy by up to 9.78 percent on benchmarks.
An adaptive, multi-tool verifier for multimodal RL that scores final answers, spatial grounding, and reasoning quality improves agentic models on spatial, hallucination, embodied, and robotic benchmarks.
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.
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.
DeFacto trains multimodal models with counterfactual image variants and GRPO reinforcement learning to enforce that correct answers are supported by correct visual evidence.
Survey that defines agentic RL for LLMs via POMDPs, introduces a taxonomy of planning/tool-use/memory/reasoning capabilities and domains, and compiles open environments from over 500 papers.
OmniView-Space framework with MPSM, tool-guided reasoning, and distillation achieves SOTA on spatial reasoning benchmarks for MLLMs while reducing external geometry dependencies.
The survey formalizes MLLM perception as a unified vision-language capability and traces its evolution via a new five-stage taxonomy while outlining future challenges.
V-Zero trains MLLMs for visual reasoning without answer labels by gating on-policy distillation trajectories using contrastive evidence from relevant versus negative image crops.
PreciseDoc is an LMM that uses synthetic document data pipelines and RL-based joint grounding-reasoning training to achieve higher precision in locating elements within text-rich document images than prior approaches.
TrOPD stabilizes on-policy distillation for LLMs with trust-region learning, outlier estimation, and off-policy guidance, outperforming prior OPD methods on reasoning and code benchmarks.
MoVT unifies different visual reasoning modes in a single model and uses the AdaVaR two-stage framework with supervised cold-start and RL via AdaGRPO to enable context-adaptive mode selection, yielding consistent gains on visual reasoning tasks.
citing papers explorer
-
Perceive-to-Reason: Decoupling Perception and Reasoning for Fine-Grained Visual Reasoning
P2R decouples perception from reasoning in VLMs via a two-stage process and PRA-GRPO alternating RL training, reporting gains such as 93.2% on V-Star for the 4B model over its Qwen3-VL backbone.
-
CaMo: Camera Motion Grounded Evaluation and Training for Vision-Language Models
Proposes Spatial Narrative Score (SNS) evaluation for VLMs' camera motion understanding and introduces CaMo model achieving consistent performance on SNS and direct QA.
-
Mind's Eye: A Benchmark of Visual Abstraction, Transformation and Composition for Multimodal LLMs
Mind's Eye benchmark shows top multimodal LLMs score below 50% on visual abstraction, relation, and transformation tasks while humans reach 80%.
-
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.
-
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.
-
Latent Noise Mask for Reducing Visual Redundancy in Multimodal Large Language Models
Lens purifies visual evidence in MLLMs via question-conditioned latent noise masking with a LET token, yielding 2.4-6.4 point gains on VQA and grounding tasks.
-
Self-Prophetic Decoding to Unlock Visual Search in LVLMs
SeProD is a plug-and-play self-prophetic decoding framework that combines pre- and post-training LVLM capabilities via probability-based sampling to improve coherent visual search and multi-step reasoning.
-
Look-Closer-Then-Diagnose: Confidence-Aware Ultrasound VQA via Active Zooming
Introduces Zoom-then-Diagnose paradigm and uncertainty-aware reward in GRPO for confidence-aware ultrasound VQA, reporting 39.3% improvement in lesion localization across liver, breast, and thyroid datasets.
-
Don't Show Pixels, Show Cues: Unlocking Visual Tool Reasoning in Language Models via Perception Programs
Perception Programs rewrite dense visual tool outputs into language-native summaries, boosting MLLM accuracy by 15-45% absolute on BLINK perception tasks and setting new state-of-the-art results.
-
Test-time Scaling over Perception: Resolving the Grounding Paradox in Thinking with Images
TTSP samples and filters multiple zoom-in exploration traces and iteratively consolidates validated observations into an Evidence Ledger, improving fine-grained multimodal reasoning on V* Bench, HR-Bench, TreeBench, and MME-RealWorld-Lite.
-
Scientific Graphics Program Synthesis via Dual Self-Consistency Reinforcement Learning
SciTikZer-8B uses a new dataset, benchmark, and dual self-consistency RL to generate TikZ code for scientific graphics, outperforming much larger models like Gemini-2.5-Pro.
-
Discovering Failure Modes in Vision-Language Models using RL
An RL-based questioner agent adaptively generates queries to discover novel failure modes in VLMs without human intervention.
-
CharTool: Tool-Integrated Visual Reasoning for Chart Understanding
CharTool equips MLLMs with cropping and code tools plus agentic RL on DuoChart data to raise chart-reasoning accuracy by up to 9.78 percent on benchmarks.
-
Multimodal Reinforcement Learning with Adaptive Verifier for AI Agents
An adaptive, multi-tool verifier for multimodal RL that scores final answers, spatial grounding, and reasoning quality improves agentic models on spatial, hallucination, embodied, and robotic benchmarks.
-
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.
-
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.
-
DeFacto: Counterfactual Thinking with Images for Enforcing Evidence-Grounded and Faithful Reasoning
DeFacto trains multimodal models with counterfactual image variants and GRPO reinforcement learning to enforce that correct answers are supported by correct visual evidence.
-
The Landscape of Agentic Reinforcement Learning for LLMs: A Survey
Survey that defines agentic RL for LLMs via POMDPs, introduces a taxonomy of planning/tool-use/memory/reasoning capabilities and domains, and compiles open environments from over 500 papers.
-
OmniView-Space: Reinforcing Spatial Reasoning via Multi-Perspective Spatial Mapping
OmniView-Space framework with MPSM, tool-guided reasoning, and distillation achieves SOTA on spatial reasoning benchmarks for MLLMs while reducing external geometry dependencies.
-
From Structure to Synergy: A Survey of Vision-Language Perception Paradigm Evolution in Multimodal Large Language Models
The survey formalizes MLLM perception as a unified vision-language capability and traces its evolution via a new five-stage taxonomy while outlining future challenges.
-
V-Zero: Answer-Label-Free On-Policy Distillation with Contrastive Evidence Gating for Fine-Grained Visual Reasoning
V-Zero trains MLLMs for visual reasoning without answer labels by gating on-policy distillation trajectories using contrastive evidence from relevant versus negative image crops.
-
An LMM for Precisely Grounding Elements in Documents
PreciseDoc is an LMM that uses synthetic document data pipelines and RL-based joint grounding-reasoning training to achieve higher precision in locating elements within text-rich document images than prior approaches.
-
Trust Region On-Policy Distillation
TrOPD stabilizes on-policy distillation for LLMs with trust-region learning, outlier estimation, and off-policy guidance, outperforming prior OPD methods on reasoning and code benchmarks.
-
Mixture-of-Visual-Thoughts: Exploring Context-Adaptive Reasoning Mode Selection for General Visual Reasoning
MoVT unifies different visual reasoning modes in a single model and uses the AdaVaR two-stage framework with supervised cold-start and RL via AdaGRPO to enable context-adaptive mode selection, yielding consistent gains on visual reasoning tasks.
-
UniReason-Med: A Shared Grounded Reasoning Interface for 2D-to-3D Transfer in Medical VQA
UniReason-Med introduces a unified framework for 2D and 3D medical VQA with shared grounded reasoning, trained on a 220K dataset, claiming that joint 2D+3D supervision improves 3D performance over 3D-only training.
-
Mags-RL: Wearing Multimodal LLMs a Magnifying Glass via Agentic Reinforcement Learning For Complex Scene Reasoning
Mags-RL uses agentic RL and a super-resolution agent for two-round reasoning in MLLMs, claiming gains on VSR, TallyQA, and GQA with a curriculum needing only 40 samples.
-
A Survey of Reinforcement Learning for Large Reasoning Models
A survey compiling RL methods, challenges, data resources, and applications for enhancing reasoning in large language models and large reasoning models since DeepSeek-R1.