Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.18858

Visual Tokens on Demand: Visual Budgeting for Scalable and Efficient Multi-Image Reasoning in Multimodal LLMs

Aimon Rahman·Joymallya Chakraborty·Vidit Bansal·Abhishek Dan
View original source ↗Read source document ↗Export metadata ↓

Abstract

Multimodal large language models have advanced rapidly, but most remain optimized for single-image inputs. Multi-image reasoning is costly because each image adds many visual tokens, increasing encoding and attention demands. Existing techniques that address the above issue, including visual token pruning, token merging, visual token compression, learned resampling, hierarchical compression, retrieval-augmented visual reasoning, and memory-based architectures, have largely been designed for single-image scenarios, where redundancy is primarily spatial and local. Multi-image reasoning introduces additional challenges: redundancy may occur across images, task-relevant evidence may be sparse or distributed, and aggressive per-image compression can remove details needed for comparison, counting, temporal reasoning, retrieval, or evidence grounding. Moreover, existing token-reduction methods are often either (i) training-free but inference-inefficient, typically relying on accessing attention matrices to get token-importance that may be impractical to access with memory-efficient attention implementations such as FlashAttention, or (ii) training-dependent, requiring the base MLLM to be retrained or adapted to a new compressed-token paradigm. In this work, we study scalable multi-image reasoning and understanding without retraining the base model and without accessing attention matrices at inference time. To this end, we propose mmVB, a lightweight plug-and-play module that assigns query-conditioned per-image token budgets via cross-image context, adding only 1.6M trainable parameters while reducing visual input tokens by 9x. This reduction yields 5x faster time-to-first-token and 7x fewer prefill FLOPs, while maintaining 94.4% of full-resolution accuracy across eight multi-image reasoning benchmarks.

Explore related subjects

cs.CV

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Aimon Rahman, Joymallya Chakraborty, Vidit Bansal, Abhishek Dan. 2026-10-13. Visual Tokens on Demand: Visual Budgeting for Scalable and Efficient Multi-Image Reasoning in Multimodal LLMs. https://arxiv.org/abs/2610.18858

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

A LoRA is Worth a Thousand Pictures

Low-Rank Adaptation (LoRA) has become a standard approach for parameter-efficient fine-tuning of image diffusion models, with over 200K models trained to replicate artistic styles and shared in online communities. One way to analyze styles of LoRA models is to generate image samples and apply image-based features like CLIP. However, this generation requires additional computation and prompt selection. In this work, we propose analyzing styles directly in the LoRA weight space, bypassing these limitations while achieving better performance even on unseen LoRAs. Since in-the-wild LoRAs lack ground truth model details, we design an experimental setup and show that LoRA weights alone serve as effective feature representations of adapted styles, robust to variations in fine-tuning seeds and image selection. We extensively evaluate using SD1.5 (the most popular base model within online communities) across data domains (fine-art paintings, architectural photos) and diverse prompt settings, with a small-scale SDXL experiment further verifying generalization. These evaluations confirm that our approach, under the selected aspects of variability, preserves fine-grained styles and outperforms image-based methods in clustering and retrieval of unseen LoRAs.

cs.CV↗
arXiv2026

Video Depth Anything: Consistent Depth Estimation for Super-Long Videos

Depth Anything has achieved remarkable success in monocular depth estimation with strong generalization ability. However, it suffers from temporal inconsistency in videos, hindering its practical applications. Various methods have been proposed to alleviate this issue by leveraging video generation models or introducing priors from optical flow and camera poses. Nonetheless, these methods are only applicable to short videos (< 10 seconds) and require a trade-off between quality and computational efficiency. We propose Video Depth Anything for high-quality, consistent depth estimation in super-long videos (over several minutes) without sacrificing efficiency. We base our model on Depth Anything V2 and replace its head with an efficient spatial-temporal head. We design a straightforward yet effective temporal consistency loss by constraining the temporal depth gradient, eliminating the need for additional geometric priors. The model is trained on a joint dataset of video depth and unlabeled images, similar to Depth Anything V2. Moreover, a novel key-frame-based strategy is developed for long video inference. Experiments show that our model can be applied to arbitrarily long videos without compromising quality, consistency, or generalization ability. Comprehensive evaluations on multiple video benchmarks demonstrate that our approach sets a new state-of-the-art in zero-shot video depth estimation. We offer models of different scales to support a range of scenarios, with our smallest model capable of real-time performance at 30 FPS.

cs.CV↗
arXiv2026

Uncertainty Propagation for Echocardiography Clinical Metric Estimation via Contour Sampling

Echocardiography plays a fundamental role in the extraction of clinical parameters (e.g. left ventricular volume and ejection fraction) required to determine the presence and severity of heart-related conditions. When deploying automated techniques for computing these parameters, uncertainty estimation is highly valuable for assessing their utility. Since clinical parameters are usually derived from segmentation maps, there is no clear path for converting pixel-wise uncertainty values into uncertainty estimates in the downstream clinical metric calculation. In this work, we propose a novel uncertainty estimation method based on contouring rather than segmentation. Our method explicitly predicts contour location uncertainty from which contour samples can be drawn. Finally, the sampled contours can be used to propagate uncertainty to clinical metrics. Our proposed method not only provides accurate uncertainty estimations for the task of contouring but also for the downstream clinical metrics on one public and two private cardiac ultrasound datasets containing a total of 1264 patients. Code is available at: https://github.com/ThierryJudge/contouring-uncertainty.

cs.CV↗