💡 The paper introduces a unified framework called Perceive-to-Reason that improves fine-grained visual reasoning performance on high-resolution images. Fine-grained visual reasoning is a challenging task for vision-language models, especially when small but critical visual cues are buried in high-resolution images. Existing approaches typically do not explicitly distinguish between perception and reasoning, instead relying on repeated cropping or test-time visual search to introduce local evidence.

The Perceive-to-Reason framework addresses this limitation by formulating fine-grained visual reasoning as a two-stage process. In the first stage, the model localizes question-relevant evidence as a Perceiver, and in the second stage, it answers the question as a Reasoner based on the annotated image and cropped regions. To train the model, the authors introduce a role-aware reinforcement learning strategy called Perception-Reasoning Alternating GRPO, which alternates between perception-focused and reasoning-focused updates using only final-answer supervision.

The Perceive-to-Reason framework is built on top of existing vision-language models, and it consistently improves performance across model scales. The results show that the Perceive-to-Reason framework achieves state-of-the-art performance on several benchmarks, including V-Star, HR-Bench-4K, and HR-Bench-8K. Specifically, the P2R-4B model achieves 93.2 percent on V-Star, 81.9 percent on HR-Bench-4K, and 80.5 percent on HR-Bench-8K, substantially outperforming its corresponding backbone.

The benefits of the Perceive-to-Reason framework extend beyond high-resolution benchmarks to broader multimodal reasoning tasks. The results suggest that explicitly decoupling perception from reasoning provides an effective framework for fine-grained visual reasoning. Overall, the paper contributes a novel framework for fine-grained visual reasoning that improves performance on high-resolution images and has broader implications for multimodal reasoning tasks.

💡 The paper introduces Representation Distribution Matching, a method for one-step visual generation that matches feature distributions under pretrained encoders. The goal is to generate high-quality images by comparing the distributions of generated and reference features. The authors identify two key design axes: how the distributions are compared and the representations they are compared in. They conduct controlled studies and find three main results.

First, they show that the Maximum Mean Discrepancy, a classical method that was previously ineffective, becomes a strong and scalable objective when estimated correctly. Second, they find that the batch size of the generated images has a significant impact on performance, with an optimum batch size above 2048, which is much larger than typical batch sizes. Third, they demonstrate that using a single representation can be gamed, resulting in low scores despite visibly fake images, and instead propose using a balanced set of encoders and evaluating with a Sliced-Wasserstein distance over 14 encoders.

The authors combine these findings to develop an improved Representation Distribution Matching method, which they call iRDM. They evaluate iRDM on the ImageNet dataset and achieve state-of-the-art results, with a Sliced-Wasserstein distance of 1.30. Additionally, they use a human-preference proxy, called PickScore, which shows that iRDM is preferred over the previous best one-step generator on 71.2% of matched samples. They also apply the same method to post-train a four-step generator, called FLUX.2, and achieve better results than the original four-step version, with improved performance on GenEval and PickScore, and requiring only 90 GPU-hours. Overall, the paper presents a new method for one-step visual generation that achieves state-of-the-art results and can be used to improve existing generators.

💡 The paper introduces a new approach to studying long-horizon large language model agents, called AgenticSTS. The problem addressed is that current methods for analyzing memory components in these agents are limited, as they append past observations and reflections to every prompt, making it hard to isolate the effect of a single memory component. To solve this, the authors propose a bounded contract approach, where every decision is made from a fresh user message assembled by typed retrieval, with no raw cross-decision transcript appended. This allows for isolated analysis of memory components and demonstrates improved performance in complex decision-making tasks.

The method involves instantiating this contract in a closed-rule stochastic deck-building game, where runs require hundreds of tactical and strategic decisions. The authors create a testbed, called AgenticSTS, which includes a reproducible environment, frozen memory and skill snapshots, prompt records, and analysis scripts. This testbed allows for the study of how explicit memory layers shape long-horizon LLM-agent decisions.

The results show that the proposed approach leads to improved performance in the game, with a fixed-A0 ablation showing the largest observed difference when triggered strategic skills are enabled. The no-store baseline wins 3 out of 10 games, while adding the skill layer wins 6 out of 10 games. Although the comparison is directional rather than statistically decisive, the results demonstrate the effectiveness of the proposed approach. The authors also release a public online benchmark of frontier LLMs on the same game, which reports zero wins at the lowest difficulty across five configurations, highlighting the challenge of the task. Overall, the paper contributes a new methodology for studying long-horizon LLM agents and demonstrates its effectiveness in a complex decision-making task.

💡 The paper proposes a training-free acceleration strategy for text-to-image diffusion models called MrFlow. The problem with existing multi-resolution generation strategies is that they can produce noticeable blurring or artifacts due to upsampling in the latent space and selective modification of partial regions. MrFlow addresses this issue by using a staged low-to-high-resolution pipeline. It first generates the main structure at low resolution, then performs super-resolution in the pixel space using a lightweight pretrained model, injects low-strength noise to enable high-frequency resampling, and finally refines the details at high resolution. The results show that MrFlow achieves a 10x end-to-end acceleration while maintaining a high level of image quality, with only a 1 percent gap in performance compared to the original model. Additionally, MrFlow can be combined with other acceleration strategies, such as timestep distillation, to achieve even higher acceleration of up to 25x. The key advantage of MrFlow is that it does not require any training or runtime modifications, making it a hardware-agnostic and efficient solution for accelerating text-to-image diffusion models.

💡 MetaSpatial is a framework that uses reinforcement learning to improve 3D spatial reasoning in vision-language models, which are used to generate 3D scenes. The problem with current models is that they lack internalized 3D spatial reasoning, which limits their ability to generate realistic layouts. Additionally, traditional supervised fine-tuning methods are not effective for layout generation tasks because perfect ground truth annotations are not available.

To address these challenges, MetaSpatial introduces a multi-turn reinforcement learning-based optimization mechanism that integrates physics-aware constraints and rendered image evaluations. This mechanism allows the model to refine spatial arrangements over multiple turns by analyzing rendered outputs, improving scene coherence progressively.

The method works by having the model analyze rendered outputs and refine the spatial arrangements in an iterative process. This process ensures that the generated 3D layouts are coherent, physically plausible, and aesthetically consistent.

The results of the empirical evaluations demonstrate that MetaSpatial significantly enhances the spatial consistency and formatting stability of various scale models. After training, object placements are more realistic, aligned, and functionally coherent, which validates the effectiveness of reinforcement learning for 3D spatial reasoning in applications such as metaverse, AR/VR, digital twins, and game development.

Overall, the contributions of MetaSpatial are the introduction of a reinforcement learning-based framework that enhances 3D spatial reasoning in vision-language models, and the demonstration of its effectiveness in generating realistic and coherent 3D scenes. The code, data, and training pipeline are publicly available, which can facilitate further research and development in this area.