Optimizing large-scale machine learning systems, such as recommendation models for global video platforms, requires navigating a massive hyperparameter search space and, more critically, designing sophisticated optimizers, architectures, and reward functions to capture nuanced user behaviors. Achieving substantial improvements in these areas is a non-trivial task, traditionally relying on extensive manual iterations to test new hypotheses. We propose a self-evolving system that leverages Large Language Models (LLMs), specifically those from Google's Gemini family, to autonomously generate, train, and deploy high-performing, complex model changes within an end-to-end automated workflow. The self-evolving system is comprised of an Offline Agent (Inner Loop) that performs high-throughput hypothesis generation using proxy metrics, and an Online Agent (Outer Loop) that validates candidates against delayed north star business metrics in live production. Our agents act as specialized Machine Learning Engineers (MLEs): they exhibit deep reasoning capabilities, discovering novel improvements in optimization algorithms and model architecture, and formulating innovative reward functions that target long-term user engagement. The effectiveness of this approach is demonstrated through several successful production launches at YouTube, confirming that autonomous, LLM-driven evolution can surpass traditional engineering workflows in both development velocity and model performance.

Image-to-text (I2T) understanding and text-to-image (T2I) generation are two fundamental, important yet traditionally isolated multimodal tasks. Despite their intrinsic connection, existing approaches typically optimize them independently, missing the opportunity for mutual enhancement. In this paper, we argue that both tasks can be connected under a shared Auto-Encoder perspective, where text serves as the intermediate latent representation, bridging the two directions — encoding images into textual semantics (I2T) and decoding text back into images (T2I). Our key insight is that if the encoder truly "understands" the image, it should capture all essential structure, and if the decoder truly "understands" the text, it should recover that structure faithfully. Building upon this principle, we propose Unified-GRPO, a post-training method based on reinforcement learning that jointly optimizes both modules through reconstructive rewards, maximizing the semantic consistency between the input and the generated images. Under this reconstruction objective, the encoder is encouraged to extract as much accurate and comprehensive semantic information from the input image to maximize reconstruction quality, while the decoder is simultaneously optimized to generate conditioned on the encoder's prior, enabling a self-evolving improvement. Empirically, we find that using text as the intermediate representation and training under a reconstructive RL paradigm effectively benefits both I2T and T2I.The I2T module gains stronger fine-grained visual perception, such as small-object recognition, grounding, etc, while its dense embeddings and language priors, in turn, provide richer semantic signals that improve T2I fidelity and complex instruction following. These results demonstrate that the reconstructive RL establishes a mutually reinforcing cross-modal synergy within the auto-encoding framework.

In the field of multimodal chain-of-thought (CoT) reasoning, existing approaches predominantly rely on reasoning on pure linguistic space, which inherently suffers from language bias and is largely confined to math or science domains. This narrow focus limits their ability to handle complex visual reasoning tasks that demand comprehensive understanding of image details. To address these limitations, this paper introduces VGR, a novel reasoning multimodal large language model (MLLM) that can replay the visual memory during thinking just like humans. Unlike traditional MLLMs, VGR first thinks the question and detects relevant regions that may help solve problems, then, the visual memory from the critical area is extracted to assist reasoning. To achieve this, we curate a large-scale SFT dataset called VGR-SFT that contains reasoning data with mixed vision grounding and language deduction. This teaches VGR to think and actively choose grounding areas for key information before answering, and we propose a dynamic visual memory replay stage to integrates the corresponding information into the reasoning process, enhancing multimodel comprehension. Experiments on the LLaVA-NeXT-7B baseline show that VGR achieves superior performance on multimodal benchmarks requiring comprehensive image detail understanding. Compared to the baseline, VGR uses only 30% of the image token count while delivering scores of +4.1 on MMStar, +7.1 on AI2D, and +12.9 improvement on ChartQA.

Open-vocabulary segmentation (OVS) allows models to segment any categories based on text prompts, overcoming the limitations of traditional closed-set methods. Open-vocabulary segmentation (OVS) in remote sensing is limited by its reliance on standard red, green, and blue (RGB) images. This prevents models from using valuable information from other spectral bands, such as near-infrared (NIR). This article proposes MovSeg, a framework for multispectral open-vocabulary segmentation. Our method efficiently integrates four-band (RGB+NIR) data into a pretrained vision-language model. MovSeg introduces two components: a multispectral input adaptation (MIA) module and a spatial-channel adaptive tuning (SCAT) strategy. The MIA module adapts the input layer to process four-band data. It uses a weight-copying strategy for initialization and a window efficient channel attention mechanism to reweight spectral channels. The SCAT strategy is a parameter-efficient strategy that uses a hybrid of low-rank adaptation and adapters to fine-tune deep features with low computational cost. Experiments on three multispectral datasets show that MovSeg outperforms existing general-domain and remote sensing OVS methods. The model achieves significant gains on NIR-sensitive classes, confirming its ability to exploit the extra spectral data. The codes will be coming soon.

Scaling Vision-Language-Action (VLA) models on large-scale data offers a promising path to achieving a more generalized driving intelligence. However, VLA models are limited by a ``supervision deficit'': the vast model capacity is supervised by sparse, low-dimensional actions, leaving much of their representational power underutilized. To remedy this, we propose DriveVLA-W0, a training paradigm that employs world modeling to predict future images. This task generates a dense, self-supervised signal that compels the model to learn the underlying dynamics of the driving environment. We showcase the paradigm's versatility by instantiating it for two dominant VLA archetypes: an autoregressive world model for VLAs that use discrete visual tokens, and a diffusion world model for those operating on continuous visual features. Building on the rich representations learned from world modeling, we introduce a lightweight action expert to address the inference latency for real-time deployment. Extensive experiments on the NAVSIM benchmark and a 700x larger in-house dataset demonstrate that DriveVLA-W0 significantly outperforms BEV and VLA baselines. Crucially, it amplifies the data scaling law, showing that performance gains accelerate as the training dataset size increases. Code will be made publicly available.

Out-of-distribution (OOD) detection represents a critical challenge in remote sensing applications, where reliable identification of novel or anomalous patterns is essential for autonomous monitoring, disaster response, and environmental assessment. Despite remarkable progress in OOD detection for natural images, existing methods and benchmarks remain poorly suited to remote sensing imagery due to data scarcity, complex multi-scale scene structures, and pronounced distribution shifts. To this end, we propose RS-OOD, a novel framework that leverages remote sensing-specific vision-language modeling to enable robust few-shot OOD detection. Our approach introduces three key innovations: spatial feature enhancement that improved scene discrimination, a dual-prompt alignment mechanism that cross-verifies scene context against fine-grained semantics for spatial-semantic consistency, and a confidence-guided self-training loop that dynamically mines pseudo-labels to expand training data without manual annotation. RS-OOD consistently outperforms existing methods across multiple remote sensing benchmarks and enables efficient adaptation with minimal labeled data, demonstrating the critical value of spatial-semantic integration.

Captions serve as a primary supervision signal for both multimodal understanding and text-to-image generation. However, previous evaluations treat the caption quality as a single scalar objective, which conflates two distinct properties: (1) how much visual information a caption covers and (2) how reliably the image supports its stated claims. To this end, we design a decoupled caption evaluation benchmark, CAPEval (Coverage And Precision Evaluation), with human-written ground-truth captions and human-verified atomic checklist items. Specifically, CAPEval decomposes caption quality into Coverage and Precision. The former quantifies how thoroughly a caption covers ground-truth factual content, while the latter reflects the factual correctness rate of all claims expressed in the caption. We select 10 captioners and further conduct controlled downstream end-to-end experiments with them from four model families, where the caption source is the only variable. Empirically, we find a consistent task-dependent dissociation: Coverage serves as the stronger correlate for understanding performance, whereas Precision acts as the dominant predictor for generation performance. This decoupled evaluation paradigm not only delivers a more fine-grained diagnosis of caption quality, but also offers actionable guidance for selecting and optimizing captioners tailored to different downstream tasks.

Recent video multimodal large language models achieve impressive results across various benchmarks. However, current evaluations suffer from two critical limitations: (1) inflated scores can mask deficiencies in fine-grained visual understanding and reasoning, and (2) answer correctness is often measured without verifying whether models identify the precise spatio-temporal evidence supporting their predictions. To address this, we present VideoZeroBench, a hierarchical benchmark designed for challenging long-video question answering that rigorously verifies spatio-temporal evidence. It comprises 500 manually annotated questions across 13 domains, paired with temporal intervals and spatial bounding boxes as evidence. To disentangle answering generation, temporal grounding, and spatial grounding, we introduce a five-level evaluation protocol that progressively tightens evidence requirements. Experiments show that even Gemini-3-Pro correctly answers fewer than 17

Long-form image captioning exposes a reward granularity problem in RL: captions are judged as whole sequences, while the important errors occur at the level of individual visual claims. A good dense caption should be both faithful and informative, avoiding hallucination without omitting salient details. Yet pairwise preferences, reference-based metrics, and holistic scalar rewards compress these local errors into a single sequence-level signal, obscuring the tradeoff between factuality and coverage. We introduce ClaimDiff-RL, a framework that uses reference-conditioned atomic claim differences as the reward unit for caption RL. Given an image, an actor caption, and a reference caption, a multimodal judge enumerates visually grounded differences, verifies each difference against the image, assigns open-vocabulary error types and severity levels, and produces per-difference statistics for reward composition. This makes hallucinated claims and omitted salient facts separately measurable and tunable. Experiments show that holistic scalar rewards can reduce hallucination by increasing missing facts, while ClaimDiff-RL exposes this faithfulness and coverage tradeoff and enables more balanced operating points. On a 160-image human-labeled diagnostic benchmark, public captioning benchmarks, and VQA benchmarks, ClaimDiff-RL improves the hallucination–missing-fact balance, preserves general capability, and even surpasses Gemini-3-Pro-Preview on several fine-grained Capability dimensions such as object counting, spatial relations, and scene recognition. These results suggest that typed, verifiable claim differences are an effective reward unit for fine-grained and diagnosable caption RL.

Can an agent learn a numerical search strategy through executable practice and then transfer that strategy as text? We study low-budget black-box optimization, where unaided language models remain well below strong classical optimizers. During development, an agent repeatedly writes and evaluates optimizer programs. It then distills the resulting program and practice record once into a 197-word primary Harness A, which is frozen before evaluation. Harness A reduces Gemini Flash regret by 48% in an independent N=30 study (p<.001), enters the GP-BO performance range on the practice family, and lowers mean regret on all three held-out BBOB landscapes. The same text improves every tested Gemini executor and transfers to Claude Sonnet, reducing regret by 43% and 49% (p≤.005). An independent end-to-end replication produces Harness B, a different program and text at the same performance tier. The same framework also attains the lowest regret on a sealed YouTube reward-tuning production benchmark. Executable practice is thus a viable way to discover a search policy, and language a portable medium for deploying it.

While Multimodal Large Language Models (MLLMs) excel at holistic understanding, they struggle with the dense world, i.e., complex scenes requiring fine-grained analysis of intricate details and object inter-relationships. Region-level MLLMs have been a promising step. However, previous attempts are generally optimized to understand given regions in isolation, neglecting crucial global contexts. To address this, we introduce Grasp Any Region (GAR) for comprehensive region-level visual understanding. Empowered by an effective RoI-aligned feature replay technique, GAR supports (1) precise perception by leveraging necessary global contexts, and (2) modeling interactions between multiple prompts. Together, it then naturally achieves (3) advanced compositional reasoning to answer specific free-form questions about any region, shifting the paradigm from passive description to active dialogue. Moreover, we construct GARBench, which not only provides a more accurate evaluation of single-region comprehension, but also, more importantly, measures interactions and complex reasoning across multiple regions. Empirically, GAR-1B not only maintains the state-of-the-art captioning capabilities, e.g., outperforming DAM-3B +4.5 on DLC-Bench, but also excels at modeling relationships between multiple prompts with advanced comprehension capabilities, even surpassing InternVL3-78B on GARBench-VQA. More importantly, our zero-shot GAR-8B even outperforms in-domain VideoRefer-7B on VideoRefer-BenchQ, indicating its strong comprehension capabilities can be easily transferred to videos. Code and data will be released to the community.

This paper explores multi-turn visual reasoning and observes that MLLMs repeatedly fail to localize the target, leading to long, redundant trajectories. We attribute this failure to the entanglement of reasoning and perception within a single model, the MLLM reasons and localizes simultaneously, and inaccurate localization triggers additional reasoning turns that bloat the trajectory. To solve this problem, we propose PixelEyes, a multi-turn visual reasoning agent that explicitly decouples reasoning from perception, i.e., the reasoner decides what to look for, while a specialized perception tool answers where it is. Specifically, PixelEyes introduces 1) Mask-guided Visual Search. A referring segmentation model is invoked to provide mask-precise localization, freeing the reasoner from the need to compensate for imprecise grounding. 2) Semantic-region Breadth-first Search (BFS). To eliminate redundant loops caused by repeatedly cropping incorrect sub-regions, we organize exploration as a breadth-first search over semantic regions. To internalize these capabilities, we construct the PixelEyes-6K dataset by resynthesizing expert trajectories from existing data. This explicitly embeds our mask-guided search and BFS logic into the model. We further introduce Pinpoint-Bench, a zero-hint visual search benchmark, i.e., no location cues are provided in the question, with instance-level masks and bounding boxes that separate localization failures from reasoning failures, enabling fine-grained analysis of failure modes such as inattentional blindness. Recent state-of-the-art MLLMs and visual reasoning agents leave large headroom on Pinpoint-Bench, demonstrating its quality and difficulty. Code and models are open-sourced.

This paper introduces Actor as Its Own Critic, a unified reinforcement learning framework, Cycle Group Relative Policy Optimization (CycleGRPO), that jointly optimizes region understanding and localization for Multimodal Large Language Models (MLLMs). Unlike existing separate pipelines, we leverage the inherent duality between the two tasks to construct a self-evaluating reinforcement learning paradigm: "region → text → region”. Specifically, a single MLLM first acts as the actor to generate region captions, then immediately transitions to a critic to ground its generated text back in the spatial domain. Therefore, CycleGRPO requires only region inputs, e.g., masks or bounding boxes, entirely bypassing the need for textual ground truths. A quality-aware token-level cycle-consistency reward is employed to assess the semantic discriminability of text captions via their physical localization accuracy. Empirically, built upon SAMTok, our CycleGRPO framework successfully bootstraps both capabilities simultaneously. Without any task-specific fine-tuning, the framework yields consistent performance gains across a wide range of benchmarks, including region captioning, region VQA, grounded dialogue, and referring segmentation. Overall, CycleGRPO offers a straightforward and scalable way to advance pixel-level capabilities in MLLMs. Code and models are released at https://github.com/devinxzhang/CycleGRPO.

Models like OpenAI-o3 pioneer visual grounded reasoning by dynamically ref- erencing visual regions, just like human “thinking with images”. However, no benchmark exists to evaluate these capabilities holistically. To bridge this gap, we propose TreeBench (Traceable Evidence Evaluation Benchmark), a diagnostic benchmark built on three principles: (1) focused visual perception of subtle targets in complex scenes, (2) traceable evidence via bounding box evaluation, and (3) second-order reasoning to test object interactions and spatial hierarchies beyond simple object localization. Prioritizing images with dense objects, we initially sample 1K high-quality images from SA-1B, and incorporate eight LMM experts to manually annotate questions, candidate options, and answers for each image. After three stages of quality control, TreeBench consists of 405 challenging vi- sual question-answering pairs, even the most advanced models struggle with this benchmark, where none of them reach 60% accuracy, e.g., OpenAI-o3 scores only 54.87. Furthermore, we introduce TreeVGR (Traceable Evidence Enhanced Visual Grounded Reasoning), a training paradigm to supervise localization and reasoning jointly with reinforcement learning, enabling accurate localizations and explainable reasoning pathways. Initialized from Qwen2.5-VL-7B, it improves V* Bench (+16.8), MME-RealWorld (+12.6), and TreeBench (+13.4), proving traceability is key to advancing vision-grounded reasoning. The code and data will be released.

We propose MotionAtlas, a system for detailed captioning of motion-centric videos, comprising (1) a dedicated human-annotated benchmark, (2) a scalable, high-quality pipeline to construct training samples, and (3) a family of powerful Video-MLLMs. Unlike conventional global motion captioning datasets, we focus on region-aware motion captioning: given a video and a spatiotemporal mask, the model generates precise descriptions of motion within the target region, thereby alleviating visual clutter and motion entanglement and enabling reliable, quantifiable evaluation. Concretely, we first build MotionAtlas-Bench, a comprehensive benchmark comprising 2,073 multiple-choice questions, meticulously annotated for a curated set of high-quality, motion-centric videos, to evaluate fine-grained motion understanding of the objects in question. Second, we design a rigorous and scalable data pipeline that leverages self-bootstrap refinement to suppress fine-grained hallucinations, yielding 159k high-quality motion captioning data. Third, we design a tailored training data composition strategy, which achieves consistent and substantial performance gains across diverse baseline Video-MLLMs, including Molmo2 and Qwen3-VL. For instance, MotionAtlas-4B surpasses Qwen3-VL-4B by an average of 5.2 percentage points across general motion benchmarks. The benchmark, dataset, and code have been released.

This paper introduces reconstructive visual instruction tuning (ROSS), a family of Large Multimodal Models (LMMs) that exploit vision-centric supervision signals. In contrast to conventional visual instruction tuning approaches that exclusively supervise text outputs, ROSS prompts LMMs to supervise visual outputs via reconstructing input images. By doing so, it capitalizes on the inherent richness and detail present within input images themselves, which are often lost in pure text supervision. However, producing meaningful feedback from natural images is challenging due to the heavy spatial redundancy of visual signals. To address this issue, ROSS employs a denoising objective to reconstruct latent representations of input images, avoiding directly regressing exact raw RGB values. This intrinsic activation design inherently encourages LMMs to maintain image detail, thereby enhancing their fine-grained comprehension capabilities and reducing hallucinations. Empirically, ROSS consistently brings significant improvements across different visual encoders and language models. In comparison with extrinsic assistance state-of-the-art alternatives that aggregate multiple visual experts, ROSS delivers competitive performance with a single SigLIP visual encoder, demonstrating the efficacy of our vision-centric supervision tailored for visual outputs.

Unified Vision-Language Models (UVLMs) perform both understanding and generation within a single architecture. Since these models rely on heterogeneous data and supervision, balancing both generation and understanding in reinforcement learning (RL) is challenging. To address this challenge, we propose PairUni, a unified framework that reorganizes data into understanding-generation (UG) pairs and aligns optimization accordingly. Specifically, we construct a unified paired dataset by synthesizing aligned instances via cross-modal semantic completion and retrieving semantically related samples. These paired structures expose cross-task semantic correspondences and support consistent policy learning. To leverage this structure, we present PairGRPO, a pair-aware variant based on Group Relative Policy Optimization. It assigns a similarity score to each pair to modulate the advantage, strengthening learning from well-aligned examples and reducing task interference. Extensive experiments across diverse UVLM architectures (Autoregressive and Discrete Diffusion) and scales (1B to 14B) demonstrate that PairUni yields consistent improvements over strong baselines. Notably, our method also demonstrates strong generalization by improving performance on image editing tasks without using any editing-specific data. Codes are available at https://github.com/Haochen-Wang409/PairUni.

Masked visual modeling has attracted much attention due to its promising potential in learning generalizable representations. Typical approaches urge models to predict specific contents of masked tokens, which can be intuitively considered as teaching a student (the model) to solve given problems (predicting masked contents). Under such settings, the performance is highly correlated with mask strategies (the difficulty of provided problems). We argue that it is equally important for the model to stand in the shoes of a teacher to produce challenging problems by itself. Intuitively, patches with high values of reconstruction loss can be regarded as hard samples, and masking those hard patches naturally becomes a demanding reconstruction task. To empower the model as a teacher, we propose Hard Patch Mining (HPM), predicting patch-wise losses and subsequently determining where to mask. Technically, we introduce an auxiliary loss predictor, which is trained with a relative objective to prevent overfitting to exact loss values. To gradually guide the training procedure, we propose an easy-to-hard mask strategy. Empirically, HPM brings significant improvements under both image and video benchmarks. Interestingly, solely incorporating the extra loss prediction objective leads to better representations, verifying the efficacy of determining where is hard to reconstruct.

With the continuous enhancement of remote sensing image resolution and the rapid advancement of deep learning techniques, land cover mapping is undergoing a significant transformation from pixel-level segmentation to object-based vector modeling. This shift imposes higher demands on deep learning models, requiring not only precise delineation of object boundaries but also the preservation of topological consistency among geographic elements. However, existing public datasets face three major limitations: limited class annotations, restricted data scale, and the lack of spatial structural information, which severely hinder the development of breakthrough methods in high-resolution remote sensing vectorization. To address these challenges, we present IRSAMap, the first global remote sensing dataset designed for large-scale, high-resolution, multifeature land cover vector mapping. This dataset offers four key advantages. First, a comprehensive element vector annotation system that includes over 1.8 million instances of ten typical natural and man-made objects, such as buildings, roads, rivers, and trees, using a unified vector annotation standard framework that ensures both semantic integrity and spatial structural accuracy. Second, an intelligent annotation workflow incorporating "manual preannotation + AI-based training and inference + manual review and correction," which enhances annotation efficiency while ensuring consistency. Third, a global coverage that spans 79 regions across six continents, representing diverse terrain types, including urban and rural areas, with a total coverage area exceeding 1000 km2. Fourth, multitask adaptability, supporting various tasks such as pixel-level land cover classification, building outline regularization extraction, road centerline extraction, and panoramic segmentation. As a fundamental resource for remote sensing intelligent interpretation, IRSAMap provides a standardized benchmark for the paradigm shift from pixels to objects, which will significantly advance the development of high-precision geographic feature automation, collaborative modeling, and other cutting-edge research directions. The dataset is of great value for applications such as global geographic information updating and digital twin construction. IRSAMap is publicly available at https://github.com/ucas-dlg/IRSAMap