MARS2 2025 Challenge on Multimodal Reasoning: Datasets, Methods, Results, Discussion, and Outlook

This paper reviews the MARS2 2025 Challenge on Multimodal Reasoning. We aim to bring together different approaches in multimodal machine learning and LLMs via a large benchmark. We hope it better allows researchers to follow the state-of-the-art in this very dynamic area. Meanwhile, a growing number of testbeds have boosted the evolution of general-purpose large language models. Thus, this year's MARS2 focuses on real-world and specialized scenarios to broaden the multimodal reasoning applications of MLLMs. Our organizing team released two tailored datasets Lens and AdsQA as test sets, which support general reasoning in 12 daily scenarios and domain-specific reasoning in advertisement videos, respectively. We evaluated 40+ baselines that include both generalist MLLMs and task-specific models, and opened up three competition tracks, i.e., Visual Grounding in Real-world Scenarios (VG-RS), Visual Question Answering with Spatial Awareness (VQA-SA), and Visual Reasoning in Creative Advertisement Videos (VR-Ads). Finally, 76 teams from the renowned academic and industrial institutions have registered and 40+ valid submissions (out of 1200+) have been included in our ranking lists. Our datasets, code sets (40+ baselines and 15+ participants' methods), and rankings are publicly available on the MARS2 workshop website and our GitHub organization page https://github.com/mars2workshop/, where our updates and announcements of upcoming events will be continuously provided.

HLSMAC: A New StarCraft Multi-Agent Challenge for High-Level Strategic Decision-Making

Benchmarks are crucial for assessing multi-agent reinforcement learning (MARL) algorithms. While StarCraft II-related environments have driven significant advances in MARL, existing benchmarks like SMAC focus primarily on micromanagement, limiting comprehensive evaluation of high-level strategic intelligence. To address this, we introduce HLSMAC, a new cooperative MARL benchmark with 12 carefully designed StarCraft II scenarios based on classical stratagems from the Thirty-Six Stratagems. Each scenario corresponds to a specific stratagem and is designed to challenge agents with diverse strategic elements, including tactical maneuvering, timing coordination, and deception, thereby opening up avenues for evaluating high-level strategic decision-making capabilities. We also propose novel metrics across multiple dimensions beyond conventional win rate, such as ability utilization and advancement efficiency, to assess agents' overall performance within the HLSMAC environment. We integrate state-of-the-art MARL algorithms and LLM-based agents with our benchmark and conduct comprehensive experiments. The results demonstrate that HLSMAC serves as a robust testbed for advancing multi-agent strategic decision-making.

Graph-Based Feature Augmentation for Predictive Tasks on Relational Datasets

Data has become a foundational asset driving innovation across domains such as finance, healthcare, and e-commerce. In these areas, predictive modeling over relational tables is commonly employed, with increasing emphasis on reducing manual effort through automated machine learning (AutoML) techniques. This raises an interesting question: can feature augmentation itself be automated and identify and utilize task-related relational signals? To address this challenge, we propose an end-to-end automated feature augmentation framework, ReCoGNN, which enhances initial datasets using features extracted from multiple relational tables to support predictive tasks. ReCoGNN first captures semantic dependencies within each table by modeling intra-table attribute relationships, enabling it to partition tables into structured, semantically coherent segments. It then constructs a heterogeneous weighted graph that represents inter-row relationships across all segments. Finally, ReCoGNN leverages message-passing graph neural networks to propagate information through the graph, guiding feature selection and augmenting the original dataset. Extensive experiments conducted on ten real-life and synthetic datasets demonstrate that ReCoGNN consistently outperforms existing methods on both classification and regression tasks.

Prompting Wireless Networks: Reinforced In-Context Learning for Power Control

To manage and optimize constantly evolving wireless networks, existing machine learning (ML)- based studies operate as black-box models, leading to increased computational costs during training and a lack of transparency in decision-making, which limits their practical applicability in wireless networks. Motivated by recent advancements in large language model (LLM)-enabled wireless networks, this paper proposes ProWin, a novel framework that leverages reinforced in-context learning to design task-specific demonstration Prompts for Wireless Network optimization, relying on the inference capabilities of LLMs without the need for dedicated model training or finetuning. The task-specific prompts are designed to incorporate natural language descriptions of the task description and formulation, enhancing interpretability and eliminating the need for specialized expertise in network optimization. We further propose a reinforced in-context learning scheme that incorporates a set of advisable examples into task-specific prompts, wherein informative examples capturing historical environment states and decisions are adaptively selected to guide current decision-making. Evaluations on a case study of base station power control showcases that the proposed ProWin outperforms reinforcement learning (RL)-based methods, highlighting the potential for next-generation future wireless network optimization.

Safe: Enhancing Mathematical Reasoning in Large Language Models via Retrospective Step-aware Formal Verification

Chain-of-Thought (CoT) prompting has become the de facto method to elicit reasoning capabilities from large language models (LLMs). However, to mitigate hallucinations in CoT that are notoriously difficult to detect, current methods such as process reward models (PRMs) or self-consistency operate as opaque boxes and do not provide checkable evidence for their judgments, possibly limiting their effectiveness. To address this issue, we draw inspiration from the idea that "the gold standard for supporting a mathematical claim is to provide a proof". We propose a retrospective, step-aware formal verification framework $Safe$. Rather than assigning arbitrary scores, we strive to articulate mathematical claims in formal mathematical language Lean 4 at each reasoning step and provide formal proofs to identify hallucinations. We evaluate our framework $Safe$ across multiple language models and various mathematical datasets, demonstrating a significant performance improvement while offering interpretable and verifiable evidence. We also propose $FormalStep$ as a benchmark for step correctness theorem proving with $30,809$ formal statements. To the best of our knowledge, our work represents the first endeavor to utilize formal mathematical language Lean 4 for verifying natural language content generated by LLMs, aligning with the reason why formal mathematical languages were created in the first place: to provide a robust foundation for hallucination-prone human-written proofs.

AdaHuman: Animatable Detailed 3D Human Generation with Compositional Multiview Diffusion

Existing methods for image-to-3D avatar generation struggle to produce highly detailed, animation-ready avatars suitable for real-world applications. We introduce AdaHuman, a novel framework that generates high-fidelity animatable 3D avatars from a single in-the-wild image. AdaHuman incorporates two key innovations: (1) A pose-conditioned 3D joint diffusion model that synthesizes consistent multi-view images in arbitrary poses alongside corresponding 3D Gaussian Splats (3DGS) reconstruction at each diffusion step; (2) A compositional 3DGS refinement module that enhances the details of local body parts through image-to-image refinement and seamlessly integrates them using a novel crop-aware camera ray map, producing a cohesive detailed 3D avatar. These components allow AdaHuman to generate highly realistic standardized A-pose avatars with minimal self-occlusion, enabling rigging and animation with any input motion. Extensive evaluation on public benchmarks and in-the-wild images demonstrates that AdaHuman significantly outperforms state-of-the-art methods in both avatar reconstruction and reposing. Code and models will be publicly available for research purposes.

GeoMan: Temporally Consistent Human Geometry Estimation using Image-to-Video Diffusion

Estimating accurate and temporally consistent 3D human geometry from videos is a challenging problem in computer vision. Existing methods, primarily optimized for single images, often suffer from temporal inconsistencies and fail to capture fine-grained dynamic details. To address these limitations, we present GeoMan, a novel architecture designed to produce accurate and temporally consistent depth and normal estimations from monocular human videos. GeoMan addresses two key challenges: the scarcity of high-quality 4D training data and the need for metric depth estimation to accurately model human size. To overcome the first challenge, GeoMan employs an image-based model to estimate depth and normals for the first frame of a video, which then conditions a video diffusion model, reframing video geometry estimation task as an image-to-video generation problem. This design offloads the heavy lifting of geometric estimation to the image model and simplifies the video model's role to focus on intricate details while using priors learned from large-scale video datasets. Consequently, GeoMan improves temporal consistency and generalizability while requiring minimal 4D training data. To address the challenge of accurate human size estimation, we introduce a root-relative depth representation that retains critical human-scale details and is easier to be estimated from monocular inputs, overcoming the limitations of traditional affine-invariant and metric depth representations. GeoMan achieves state-of-the-art performance in both qualitative and quantitative evaluations, demonstrating its effectiveness in overcoming longstanding challenges in 3D human geometry estimation from videos.

Understanding 6G through Language Models: A Case Study on LLM-aided Structured Entity Extraction in Telecom Domain

Knowledge understanding is a foundational part of envisioned 6G networks to advance network intelligence and AI-native network architectures. In this paradigm, information extraction plays a pivotal role in transforming fragmented telecom knowledge into well-structured formats, empowering diverse AI models to better understand network terminologies. This work proposes a novel language model-based information extraction technique, aiming to extract structured entities from the telecom context. The proposed telecom structured entity extraction (TeleSEE) technique applies a token-efficient representation method to predict entity types and attribute keys, aiming to save the number of output tokens and improve prediction accuracy. Meanwhile, TeleSEE involves a hierarchical parallel decoding method, improving the standard encoder-decoder architecture by integrating additional prompting and decoding strategies into entity extraction tasks. In addition, to better evaluate the performance of the proposed technique in the telecom domain, we further designed a dataset named 6GTech, including 2390 sentences and 23747 words from more than 100 6G-related technical publications. Finally, the experiment shows that the proposed TeleSEE method achieves higher accuracy than other baseline techniques, and also presents 5 to 9 times higher sample processing speed.

VideoRFT: Incentivizing Video Reasoning Capability in MLLMs via Reinforced Fine-Tuning

Reinforcement fine-tuning (RFT) has shown great promise in achieving humanlevel reasoning capabilities of Large Language Models (LLMs), and has recently been extended to MLLMs. Nevertheless, reasoning about videos, which is a fundamental aspect of human intelligence, remains a persistent challenge due to the complex logic, temporal and causal structures inherent in video data. To fill this gap, we propose VIDEORFT, a novel approach that extends the RFT paradigm to cultivate human-like video reasoning capabilities in MLLMs. VIDEORFT follows the standard two-stage scheme in RFT: supervised fine-tuning (SFT) with chain-of-thought (CoT) annotations, followed by reinforcement learning (RL) to improve generalization. A central challenge to achieve this in the video domain lies in the scarcity of large-scale, high-quality video CoT datasets. We address this by building a fully automatic CoT curation pipeline. First, we devise a cognitioninspired prompting strategy to elicit a reasoning LLM to generate preliminary CoTs based solely on rich, structured, and literal representations of video content. Subsequently, these CoTs are revised by a visual-language model conditioned on the actual video, ensuring visual consistency and reducing visual hallucinations. This pipeline results in two new datasets - VideoRFT-CoT-102K for SFT and VideoRFT-RL-310K for RL. To further strength the RL phase, we introduce a novel semantic-consistency reward that explicitly promotes the alignment between textual reasoning with visual evidence. This reward encourages the model to produce coherent, context-aware reasoning outputs grounded in visual input. Extensive experiments show that VIDEORFT achieves state-of-the-art performance on six video reasoning benchmarks.

Not All Documents Are What You Need for Extracting Instruction Tuning Data

Instruction tuning improves the performance of large language models (LLMs), but it heavily relies on high-quality training data. Recently, LLMs have been used to synthesize instruction data using seed question-answer (QA) pairs. However, these synthesized instructions often lack diversity and tend to be similar to the input seeds, limiting their applicability in real-world scenarios. To address this, we propose extracting instruction tuning data from web corpora that contain rich and diverse knowledge. A naive solution is to retrieve domain-specific documents and extract all QA pairs from them, but this faces two key challenges: (1) extracting all QA pairs using LLMs is prohibitively expensive, and (2) many extracted QA pairs may be irrelevant to the downstream tasks, potentially degrading model performance. To tackle these issues, we introduce EQUAL, an effective and scalable data extraction framework that iteratively alternates between document selection and high-quality QA pair extraction to enhance instruction tuning. EQUAL first clusters the document corpus based on embeddings derived from contrastive learning, then uses a multi-armed bandit strategy to efficiently identify clusters that are likely to contain valuable QA pairs. This iterative approach significantly reduces computational cost while boosting model performance. Experiments on AutoMathText and StackOverflow across four downstream tasks show that EQUAL reduces computational costs by 5-10x and improves accuracy by 2.5 percent on LLaMA-3.1-8B and Mistral-7B