arXiv subjects

Vicente Ordonez

Publications and source records attributed to Vicente Ordonez.

At least 19 recordsLinked to original sources

arXiv2026

SoftmaxGRPO: Learning to Reason using Softmax Advantage Group Estimation

Group-based reinforcement learning objectives such as GRPO can allocate learning signal poorly across prompt difficulty: under binary rewards, group normalization induces a divergent weighting on easy prompts. We introduce Softmax Advantage Group Estimation (SoftmaxGRPO), a drop-in alternative that replaces z-score-normalized group advantages with temperature-scaled softmax advantages, keeping weights bounded regardless of prompt difficulty. For binary rewards, we derive the exact finite-group population objective and identify MaxRL as its low-temperature limit. For bounded scalar rewards, we show that the large-group update exactly optimizes a log-moment-generating-function objective, while a universal finite-group scalar objective cannot exist without additional assumptions on the reward distribution. Empirically, SoftmaxGRPO reallocates measured gradient budget away from near-solved prompts and consistently improves over GRPO under identical rewards. It reaches 51.8% on DeepMath with verifiable rewards and improves a 1.5B instruction-tuned model from 35.0% to 68.0% on Poetry using only lightweight text-similarity rewards.

cs.LG↗

arXiv2026

JigShape: Evaluating Visual-Geometric Reasoning in VLMs through Jigsaw Puzzles

Jigsaw puzzle solving requires jointly reasoning about visual content and geometric constraints, yet existing benchmarks use rectangular cuts that create ambiguous ground truth in texture-repeated regions. We introduce \textit{\ours{}}, a benchmark with tab-and-blank interlocking pieces where geometric constraints provide strong local compatibility requirements that, combined with visual content, yield unambiguous ground truth. Across 95K instances at four grid densities (4$\times$4 to 16$\times$16), we find that \textbf{zero-shot VLMs largely lack geometric reasoning}: only one of five frontier models (GPT-5.5) exceeds random baseline on 4$\times$4 puzzles, while all others perform at chance level. While supervised fine-tuning achieves $>$97\% on 4$\times$4, \textbf{all models collapse on larger grids}: GPT-5.5 drops from 70\% to near-random on 8$\times$8, and even fine-tuned models fall below 5\% on 12$\times$12. This ``scaling cliff'' suggests current architectures cannot maintain consistent constraint satisfaction as the number of pieces increases. \ours{} establishes scalable geometric reasoning as an open challenge for vision-language models.

cs.CV↗

arXiv2026

Perception Before Reasoning: Dynamic Latent Reasoning for Video Understanding and Question Answering

Video question answering requires models to ground language queries in visual evidence and, when necessary, reason over that evidence across time. Existing methods typically rely on long textual chain-of-thought rationales, even though many questions can be answered as soon as the relevant object, action, or frame is localized. We propose Dynamic Latent Reasoning (DyLaR), which first grounds a question in a short block of perception latents (continuous hidden states that encode query-relevant visual evidence), and then adaptively decides whether to append reasoning latents (continuous thoughts that reason over this evidence in latent space) before answering. DyLaR learns this behavior by grounding perception latents in verified visual evidence and distilling verified rationales into reasoning latents, followed by reinforcement learning that further refines when to reason. Across nine video benchmarks and four multimodal language model backbones, DyLaR improves average accuracy over same-backbone baselines while generating fewer than 20 tokens per query. On Qwen3-VL-4B, for example, DyLaR improves average accuracy over Qwen3-VL-4B-Thinking from 54.0 to 58.2 while reducing response length from 1,220.7 to 18.5 tokens per query. Ablations further show that grounded perception latents, rationale-supervised reasoning latents, and adaptive routing each improve accuracy.

cs.CV↗

arXiv2026

Learning to Reason by Analogy via Retrieval-Augmented Reinforcement Fine-Tuning

Retrieval-augmented generation (RAG) has become a standard mechanism for grounding language models in external knowledge, yet conventional retrieval based on lexical or semantic similarity is poorly suited for complex reasoning tasks: a semantically similar problem may demand an entirely different solution strategy, while a superficially different problem may share the same underlying reasoning pattern. We propose Retrieval-Augmented Reinforcement Fine-Tuning (RA-RFT), a post-training framework that teaches language models to reason by analogy. RA-RFT uses gold-relevance distillation to train a retriever that ranks contexts by expected reasoning benefit rather than semantic overlap, and then fine-tunes the policy model via reinforcement fine-tuning methods with retrieved analogous demonstrations, so the model learns to leverage reasoning traces under verifiable outcome rewards. We further analyze the diversity of retrieved contexts and find that reasoning-aware retrieval surfaces complementary solution strategies that provide distinct reasoning scaffolds for individual problems. Across challenging mathematical reasoning benchmarks, RA-RFT consistently outperforms standard reinforcement fine-tuning methods. For example, it improves AIME 2025 average@32 accuracy by 7.1 and 2.8 points over GRPO for Qwen3-1.7B and Qwen3-4B respectively -- suggesting that reasoning-aware retrieval is a complementary axis of improvement and orthogonal to advances in reward design or training curricula.

cs.CL↗

arXiv2026

NoiseShift: Resolution-Aware Noise Recalibration for Better Low-Resolution Image Generation

Text-to-image diffusion models often degrade when sampled at resolutions outside the final training resolution set. Prior work has largely emphasized higher resolution generation, enabling pretrained diffusion models to extrapolate beyond the resolutions seen during training. In this work, we instead target lower-resolution generation, performing inference at reduced resolution to significantly cut computational cost. We show that network conditioning of the noise level induces a train-test mismatch that directly degrades low-resolution generation: the same scheduled noise level can correspond to a different perceptual corruption level at lower resolutions, mis-calibrating the denoiser timestep and noise embedding. To this end, we propose NoiseShift, a training-free recalibration method that keeps the original noise sampling schedule unchanged and instead re-indexes the noise conditioning of the denoiser to restore local forward-reverse consistency. Using a lightweight coarse-to-fine calibration on a small set of image-text pairs, NoiseShift learns a resolution-specific mapping from scheduler noise to conditioning noise, reducing train-test mismatch and improving lower-resolution generation quality. When NoiseShift is applied to Stable Diffusion 3 (SD3), Stable Diffusion 3.5 (SD3.5), and Flux-Dev, generation quality at low resolutions improves consistently. Particularly, SD3 generation at 128x128 resolution gets an improved FID score from 203 to 171, and SD3.5 gets an improved FID score from 310 to 277 on LAION-COCO. Even Flux-Dev which already implements a complementary time-shifting strategy gets a modest boost from NoiseShift with an improved FID score from 120 to 113 at 64x64 resolution. More importantly, NoiseShift achieves such improvements with minimal implementation changes and no additional inference overhead.

cs.CV↗

arXiv2026

EntityBench: Towards Entity-Consistent Long-Range Multi-Shot Video Generation

Multi-shot video generation extends single-shot generation to coherent visual narratives, yet maintaining consistent characters, objects, and locations across shots remains a challenge over long sequences. Existing evaluations typically use independently generated prompt sets with limited entity coverage and simple consistency metrics, making standardized comparison difficult. We introduce EntityBench, a benchmark of 140 episodes (2,491 shots) derived from real narrative media, with explicit per-shot entity schedules tracking characters, objects, and locations simultaneously across easy / medium / hard tiers of up to 50 shots, 13 cross-shot characters, 8 cross-shot locations, 22 cross-shot objects, and recurrence gaps spanning up to 48 shots. It is paired with a three-pillar evaluation suite that disentangles intra-shot quality, prompt-following alignment, and cross-shot consistency, with a fidelity gate that admits only accurate entity appearances into cross-shot scoring. As a baseline, we propose EntityMem, a memory-augmented generation system that stores verified per-entity visual references in a persistent memory bank before generation begins. Experiments show that cross-shot entity consistency degrades sharply with recurrence distance in existing methods, and that explicit per-entity memory yields the highest character fidelity (Cohen's d = +2.33) and presence among methods evaluated. Code and data are available at https://github.com/Catherine-R-He/EntityBench/.

cs.CV↗

arXiv2026

Agentic Discovery with Active Hypothesis Exploration for Visual Recognition

We introduce HypoExplore, an agentic framework that formulates neural architecture discovery for visual recognition as a hypothesis-driven scientific inquiry. Given a human-specified high-level research direction, HypoExplore ideates, implements, evaluates, and improves neural architectures through evolutionary branching. New hypotheses are created using a large language model by selecting a parent hypothesis to build upon, guided by a dual strategy that balances exploiting validated principles with resolving uncertain ones. Our proposed framework maintains a Trajectory Tree that records the lineage of all proposed architectures, and a Hypothesis Memory Bank that actively tracks confidence scores acquired through experimental evidence. After each experiment, multiple feedback agents analyze the results from different perspectives and consolidate their findings into hypothesis confidence updates. Our framework is tested on discovering lightweight vision architectures on CIFAR-10, with the best achieving 94.11% accuracy evolved from a root node baseline that starts at 18.91%, and generalizes to CIFAR-100 and Tiny-ImageNet. We further demonstrate applicability to a specialized domain by conducting independent architecture discovery runs on MedMNIST, which yield a state-of-the-art performance. We show that hypothesis confidence scores grow increasingly predictive as evidence accumulates, and that the learned principles transfer across independent evolutionary lineages, suggesting that HypoExplore not only discovers stronger architectures, but can help build a genuine understanding of the design space.

cs.CV↗

arXiv2026

MetaEmbed: Scaling Multimodal Retrieval at Test-Time with Flexible Late Interaction

Universal multimodal embedding models have achieved great success in capturing semantic relevance between queries and candidates. However, current methods either condense queries and candidates into a single vector, potentially limiting the expressiveness for fine-grained information, or produce too many vectors that are prohibitive for multi-vector retrieval. In this work, we introduce MetaEmbed, a new framework for multimodal retrieval that rethinks how multimodal embeddings are constructed and interacted with at scale. During training, a fixed number of learnable Meta Tokens are appended to the input sequence. At test-time, their last-layer contextualized representations serve as compact yet expressive multi-vector embeddings. Through the proposed Matryoshka Multi-Vector Retrieval training, MetaEmbed learns to organize information by granularity across multiple vectors. As a result, we enable test-time scaling in multimodal retrieval where users can balance retrieval quality against efficiency demands by selecting the number of tokens used for indexing and retrieval interactions. Extensive evaluations on the Massive Multimodal Embedding Benchmark (MMEB) and the Visual Document Retrieval Benchmark (ViDoRe) confirm that MetaEmbed achieves state-of-the-art retrieval performance while scaling robustly to models with 32B parameters. Code is available at https://github.com/facebookresearch/MetaEmbed.

cs.IR↗

arXiv2026

Beyond Referring Expressions: Scenario Comprehension Visual Grounding

Existing visual grounding benchmarks primarily evaluate alignment between image regions and literal referring expressions, where models can often succeed by matching a prominent named category. We explore a complementary and more challenging setting of scenario-based visual grounding, where the target must be inferred from roles, intentions, and relational context rather than explicit naming. We introduce Referring Scenario Comprehension (RSC), a benchmark designed for this setting. The queries in this benchmark are paragraph-length texts that describe object roles, user goals, and contextual cues, including deliberate references to distractor objects that often require deep understanding to resolve. Each instance is annotated with interpretable difficulty tags for uniqueness, clutter, size, overlap, and position which expose distinct failure modes and support fine-grained analysis. RSC contains approximately 31k training examples, 4k in-domain test examples, and a 3k out-of-distribution split with unseen object categories. We further propose ScenGround, a curriculum reasoning method serving as a reference point for this setting, combining supervised warm-starting with difficulty-aware reinforcement learning. Experiments show that scenario-based queries expose systematic failures in current models that standard benchmarks do not reveal, and that curriculum training improves performance on challenging slices and transfers to standard benchmarks.

cs.CV↗

arXiv2026

One Model, Many Budgets: Elastic Latent Interfaces for Diffusion Transformers

Diffusion transformers (DiTs) achieve high generative quality but lock FLOPs to image resolution, limiting principled latency-quality trade-offs, and allocate computation uniformly across input spatial tokens, wasting resource allocation to unimportant regions. We introduce Elastic Latent Interface Transformer (ELIT), a drop-in, DiT-compatible mechanism that decouples input image size from compute. Our approach inserts a latent interface, a learnable variable-length token sequence on which standard transformer blocks can operate. Lightweight Read and Write cross-attention layers move information between spatial tokens and latents and prioritize important input regions. By training with random dropping of tail latents, ELIT learns to produce importance-ordered representations with earlier latents capturing global structure while later ones contain information to refine details. At inference, the number of latents can be dynamically adjusted to match compute constraints. ELIT is deliberately minimal, adding two cross-attention layers while leaving the rectified flow objective and the DiT stack unchanged. Across datasets and architectures (DiT, U-ViT, HDiT, MM-DiT), ELIT delivers consistent gains. On ImageNet-1K 512px, ELIT delivers an average gain of $35.3\%$ and $39.6\%$ in FID and FDD scores. Project page: https://snap-research.github.io/elit/

cs.CV↗

arXiv2026

DeepSport: A Multimodal Large Language Model for Comprehensive Sports Video Reasoning via Agentic Reinforcement Learning

Sports video understanding requires perceiving high-speed dynamics, complex rules, and long temporal contexts. Yet, current Multimodal Large Language Models (MLLMs) remain narrowly focused on single sports, specific tasks, or training-free paradigms. We introduce DeepSport, the first end-to-end trained MLLM for multi-task, multi-sport video understanding. DeepSport shifts from passive frame processing to active, iterative reasoning, dynamically extracting frames to "think with videos." To train our model, we curate a unified 78k-sample dataset via a rigorous three-step text-and-vision distillation pipeline. We then employ a progressive two-stage training strategy: a Sports Curriculum Supervised Fine-Tuning phase to build foundational perception, followed by Agentic Reinforcement Learning with a novel tool-use reward. Extensive experiments on a comprehensive 6.7k benchmark demonstrate that DeepSport achieves state-of-the-art performance, outperforming powerful proprietary and open-source models, while utilizing significantly fewer frames. Furthermore, it exhibits strong zero-shot transferability to unseen sports and broad motion recognition tasks, establishing a highly efficient and generalized foundation for complex video reasoning.

cs.CV↗

arXiv2026

MotionBits: Video Segmentation through Motion-Level Analysis of Rigid Bodies

Rigid bodies constitute the smallest manipulable elements in the real world, and understanding how they physically interact is fundamental to embodied reasoning and robotic manipulation. Thus, accurate detection, segmentation, and tracking of moving rigid bodies is essential for enabling reasoning modules to interpret and act in diverse environments. However, current segmentation models trained on semantic grouping are limited in their ability to provide meaningful interaction-level cues for completing embodied tasks. To address this gap, we introduce MotionBit, a novel concept that, unlike prior formulations, defines the smallest unit in motion-based segmentation through kinematic spatial twist equivalence, independent of semantics. In this paper, we contribute (1) the MotionBit concept and definition, (2) a hand-labeled benchmark, called MoRiBo, for evaluating moving rigid-body segmentation across robotic manipulation and human-in-the-wild videos, and (3) a learning-free graph-based MotionBits segmentation method that outperforms state-of-the-art embodied perception methods by 37.3\% in macro-averaged mIoU on the MoRiBo benchmark. Finally, we demonstrate the effectiveness of MotionBits segmentation for downstream embodied reasoning and manipulation tasks, highlighting its importance as a fundamental primitive for understanding physical interactions.

cs.CV↗

arXiv2026

SportR: A Benchmark for Multimodal Large Language Model Reasoning in Sports

Deeply understanding sports requires an intricate blend of fine-grained visual perception and rule-based reasoning - a challenge that pushes the limits of current multimodal models. To succeed, models must master three critical capabilities: perceiving nuanced visual details, applying abstract sport rule knowledge, and grounding that knowledge in specific visual evidence. Current sports benchmarks either cover single sports or lack the detailed reasoning chains and precise visual grounding needed to robustly evaluate these core capabilities in a multi-sport context. To address this gap, we introduce SportR, the first multi-sports large-scale benchmark designed to train and evaluate MLLMs on the fundamental reasoning required for sports intelligence. Our benchmark provides a dataset of 4,789 images and 2,052 videos. To enable granular evaluation, we structure our benchmark around a progressive hierarchy of question-answer pairs designed to probe reasoning at increasing depths - from simple infraction identification to complex penalty prediction. For the most advanced tasks requiring multi-step reasoning, such as determining penalties or explaining tactics, we provide 6,841 high-quality, human-authored Chain of Thought annotations. In addition, our benchmark incorporates both image and video modalities and provides manual bounding box annotations to test visual grounding in the image part directly. Extensive experiments demonstrate the profound difficulty of our benchmark. State-of-the-art baseline models perform poorly on our most challenging tasks. While training on our data via Supervised Fine-Tuning and Reinforcement Learning improves these scores, they remain relatively low, highlighting a significant gap in current model capabilities. SportR presents a new challenge for the community, providing a critical resource to drive future research in multimodal sports reasoning.

cs.CV↗

arXiv2025

ProxyThinker: Test-Time Guidance through Small Visual Reasoners

Recent advancements in reinforcement learning with verifiable rewards have pushed the boundaries of the visual reasoning capabilities in large vision-language models (LVLMs). However, training LVLMs with reinforcement fine-tuning (RFT) is computationally expensive, posing a significant challenge to scaling model size. In this work, we propose ProxyThinker, an inference-time technique that enables large models to inherit the visual reasoning capabilities from small, slow-thinking visual reasoners without any training. By subtracting the output distributions of base models from those of RFT reasoners, ProxyThinker modifies the decoding dynamics and successfully elicits the slow-thinking reasoning demonstrated by the emerged sophisticated behaviors such as self-verification and self-correction. ProxyThinker consistently boosts performance on challenging visual benchmarks on spatial, mathematical, and multi-disciplinary reasoning, enabling untuned base models to compete with the performance of their full-scale RFT counterparts. Furthermore, our implementation efficiently coordinates multiple language models with parallelism techniques and achieves up to 38 $\times$ faster inference compared to previous decoding-time methods, paving the way for the practical deployment of ProxyThinker. Code is available at https://github.com/MrZilinXiao/ProxyThinker.

cs.CV↗

arXiv2025

Improving Large Vision and Language Models by Learning from a Panel of Peers

Traditional alignment methods for Large Vision and Language Models (LVLMs) primarily rely on human-curated preference data. Human-generated preference data is costly; machine-generated preference data is limited in quality; and self-supervised preference data often introduces hallucinations. To overcome these limitations, we propose a novel Panel-of-Peers learning framework inspired by collaborative learning among humans. This approach leverages a panel of LVLMs, each evaluating and learning from their collective outputs through an iterative self-improvement process. By simulating a peer review system, our models generate, assess, and refine outputs in response to a curated set of prompts, mimicking a classroom learning environment. We demonstrate that this methodology enhances model performance without requiring extensive human-labeled datasets. Our experiments show significant improvement across multiple benchmarks, demonstrating the potential of peer evaluations as a scalable alternative to self-supervised alignment. Notably, we show that Panel-of-Peers increases the average score on fifteen benchmarks from 48% to 57%

cs.CV↗

arXiv2025

Evaluating Text-to-Image and Text-to-Video Synthesis with a Conditional Fréchet Distance

Evaluating text-to-image and text-to-video models is challenging due to a fundamental disconnect: established metrics fail to jointly measure visual quality and semantic alignment with text, leading to a poor correlation with human judgments. To address this critical issue, we propose cFreD, a general metric based on a Conditional Fréchet Distance that unifies the assessment of visual fidelity and text-prompt consistency into a single score. Existing metrics such as Fréchet Inception Distance (FID) capture image quality but ignore text conditioning while alignment scores such as CLIPScore are insensitive to visual quality. Furthermore, learned preference models require constant retraining and are unlikely to generalize to novel architectures or out-of-distribution prompts. Through extensive experiments across multiple recently proposed text-to-image models and diverse prompt datasets, cFreD exhibits a higher correlation with human judgments compared to statistical metrics , including metrics trained with human preferences. Our findings validate cFreD as a robust, future-proof metric for the systematic evaluation of text conditioned models, standardizing benchmarking in this rapidly evolving field. We release our evaluation toolkit and benchmark.

cs.CV↗

arXiv2025

GViT: Representing Images as Gaussians for Visual Recognition

We introduce GVIT, a classification framework that abandons conventional pixel or patch grid input representations in favor of a compact set of learnable 2D Gaussians. Each image is encoded as a few hundred Gaussians whose positions, scales, orientations, colors, and opacities are optimized jointly with a ViT classifier trained on top of these representations. We reuse the classifier gradients as constructive guidance, steering the Gaussians toward class-salient regions while a differentiable renderer optimizes an image reconstruction loss. We demonstrate that by 2D Gaussian input representations coupled with our GVIT guidance, using a relatively standard ViT architecture, closely matches the performance of a traditional patch-based ViT, reaching a 76.9% top-1 accuracy on Imagenet-1k using a ViT-B architecture.

cs.CV↗

arXiv2025

Improving Progressive Generation with Decomposable Flow Matching

Generating high-dimensional visual modalities is a computationally intensive task. A common solution is progressive generation, where the outputs are synthesized in a coarse-to-fine spectral autoregressive manner. While diffusion models benefit from the coarse-to-fine nature of denoising, explicit multi-stage architectures are rarely adopted. These architectures have increased the complexity of the overall approach, introducing the need for a custom diffusion formulation, decomposition-dependent stage transitions, add-hoc samplers, or a model cascade. Our contribution, Decomposable Flow Matching (DFM), is a simple and effective framework for the progressive generation of visual media. DFM applies Flow Matching independently at each level of a user-defined multi-scale representation (such as Laplacian pyramid). As shown by our experiments, our approach improves visual quality for both images and videos, featuring superior results compared to prior multistage frameworks. On Imagenet-1k 512px, DFM achieves 35.2% improvements in FDD scores over the base architecture and 26.4% over the best-performing baseline, under the same training compute. When applied to finetuning of large models, such as FLUX, DFM shows faster convergence speed to the training distribution. Crucially, all these advantages are achieved with a single model, architectural simplicity, and minimal modifications to existing training pipelines.

cs.CV↗