Recent works show that image and video generators exhibit zero-shot visual understanding behaviors, in a way reminiscent of how LLMs develop emergent capabilities of language understanding and reasoning from generative pretraining. While it has long been conjectured that the ability to create visual content implies an ability to understand it, there has been limited evidence that generative vision models have developed strong understanding capabilities. In this work, we demonstrate that image generation training serves a role similar to LLM pretraining, and lets models learn powerful and general visual representations that enable SOTA performance on various vision tasks. We introduce Vision Banana, a generalist model built by instruction-tuning Nano Banana Pro (NBP) on a mixture of its original training data alongside a small amount of vision task data. By parameterizing the output space of vision tasks as RGB images, we seamlessly reframe perception as image generation. Our generalist model, Vision Banana, achieves SOTA results on a variety of vision tasks involving both 2D and 3D understanding, beating or rivaling zero-shot domain-specialists, including Segment Anything Model 3 o
While vision models are highly capable, their internal mechanisms remain poorly understood -- a challenge which sparse autoencoders (SAEs) have helped address in language, but which remains underexplored in vision. We address this gap by training SAEs on CLIP's vision transformer and uncover key differences between vision and language processing, including distinct sparsity patterns for SAEs trained across layers and token types. We then provide the first systematic analysis on the steerability of CLIP's vision transformer by introducing metrics to quantify how precisely SAE features can be steered to affect the model's output. We find that 10-15\% of neurons and features are steerable, with SAEs providing thousands more steerable features than the base model. Through targeted suppression of SAE features, we then demonstrate improved performance on three vision disentanglement tasks (CelebA, Waterbirds, and typographic attacks), finding optimal disentanglement in middle model layers, and achieving state-of-the-art performance on defense against typographic attacks.
Large vision--language models (VLMs) often use a frozen vision backbone, whose image features are mapped into a large language model through a lightweight connector. While transformer-based encoders are the standard visual backbone, we ask whether state space model (SSM) vision backbones can be a strong alternative. We systematically evaluate SSM vision backbones for VLMs in a controlled setting. Under matched ImageNet-1K initialization, the SSM backbone achieves the strongest overall performance across both VQA and grounding/localization. We further adapt both SSM and ViT-family backbones with detection or segmentation training and find that dense-task tuning generally improves performance across families; after this adaptation, the SSM backbone remains competitive while operating at a substantially smaller model scale. We further observe that (i) higher ImageNet accuracy or larger backbones do not reliably translate into better VLM performance, and (ii) some visual backbones are unstable in localization. Based on these findings, we propose stabilization strategies that improve robustness for both backbone families and highlight SSM backbones as a strong alternative to transformer
Large Vision-Language Models (LVLMs) typically follow a two-stage training paradigm-pretraining and supervised fine-tuning. Recently, preference optimization, derived from the language domain, has emerged as an effective post-training reinforcement strategy to enhance capabilities of LVLMs. However, constructing high-quality human-annotated preference data and developing robust reward models to mimic these preferences are both costly and challenging. Motivated by this observation, we propose Vision-R1, a novel vision-guided R1-like reinforcement learning algorithm for LVLMs that rewards models with definitive vision feedback. It only leverages curated instruction data, eliminating the need for specialized reward models and handcrafted preference datasets. We incorporate a criterion-driven reward function that further integrates multi-dimensional feedback to evaluate model completions comprehensively based on the vision task logic. Furthermore, we introduce a progressive rule refinement strategy that dynamically adjusts the reward criteria during training, enabling continuous model improvement and mitigating reward hacking. Extensive experiments on both in-distribution and out-of-di
Robust tooling and publicly available pre-trained models have helped drive recent advances in mechanistic interpretability for language models. However, similar progress in vision mechanistic interpretability has been hindered by the lack of accessible frameworks and pre-trained weights. We present Prisma (Access the codebase here: https://github.com/Prisma-Multimodal/ViT-Prisma), an open-source framework designed to accelerate vision mechanistic interpretability research, providing a unified toolkit for accessing 75+ vision and video transformers; support for sparse autoencoder (SAE), transcoder, and crosscoder training; a suite of 80+ pre-trained SAE weights; activation caching, circuit analysis tools, and visualization tools; and educational resources. Our analysis reveals surprising findings, including that effective vision SAEs can exhibit substantially lower sparsity patterns than language SAEs, and that in some instances, SAE reconstructions can decrease model loss. Prisma enables new research directions for understanding vision model internals while lowering barriers to entry in this emerging field.
We introduce Granite Vision, a lightweight large language model with vision capabilities, specifically designed to excel in enterprise use cases, particularly in visual document understanding. Our model is trained on a comprehensive instruction-following dataset, including document-related tasks, such as content extraction from tables, charts, diagrams, sketches, and infographics, as well as general image tasks. The architecture of Granite Vision is centered around visual modality alignment with a decoder-only, 2 billion parameter Granite large language model. Additionally, we introduce a dedicated safety classification approach in test-time that leverages a sparse set of attention vectors to identify potential harmful inputs. Despite its lightweight architecture, Granite Vision achieves strong results in standard benchmarks related to visual document understanding, as well as on the LiveXiv benchmark, which is designed to avoid test set contamination by using a constantly updated corpus of recently published Arxiv papers. We are releasing the model under the Apache-2 license, allowing for both research and commercial use, while offering complete visibility into the training data a
Building state-of-the-art Vision-Language Models (VLMs) with strong captioning capabilities typically necessitates training on billions of high-quality image-text pairs, requiring millions of GPU hours. This paper introduces the Vision-Language-Vision (VLV) auto-encoder framework, which strategically leverages key pretrained components: a vision encoder, the decoder of a Text-to-Image (T2I) diffusion model, and subsequently, a Large Language Model (LLM). Specifically, we establish an information bottleneck by regularizing the language representation space, achieved through freezing the pretrained T2I diffusion decoder. Our VLV pipeline effectively distills knowledge from the text-conditioned diffusion model using continuous embeddings, demonstrating comprehensive semantic understanding via high-quality reconstructions. Furthermore, by fine-tuning a pretrained LLM to decode the intermediate language representations into detailed descriptions, we construct a state-of-the-art (SoTA) captioner comparable to leading models like GPT-4o and Gemini 2.0 Flash. Our method demonstrates exceptional cost-efficiency and significantly reduces data requirements; by primarily utilizing single-modal
Building on recent advances in language-based reasoning models, we explore multimodal reasoning that integrates vision and text. Existing multimodal benchmarks primarily test visual extraction combined with text-based reasoning, lacking true visual reasoning with more complex interactions between vision and language. Inspired by the ARC challenge, we introduce EasyARC, a vision-language benchmark requiring multi-image, multi-step reasoning, and self-correction. EasyARC is procedurally generated, fully verifiable, and scalable, making it ideal for reinforcement learning (RL) pipelines. The generators incorporate progressive difficulty levels, enabling structured evaluation across task types and complexities. We benchmark state-of-the-art vision-language models and analyze their failure modes. We argue that EasyARC sets a new standard for evaluating true reasoning and test-time scaling capabilities in vision-language models. We open-source our benchmark dataset and evaluation code.
Recently, we have witnessed the great success of the generalist model in natural language processing. The generalist model is a general framework trained with massive data and is able to process various downstream tasks simultaneously. Encouraged by their impressive performance, an increasing number of researchers are venturing into the realm of applying these models to computer vision tasks. However, the inputs and outputs of vision tasks are more diverse, and it is difficult to summarize them as a unified representation. In this paper, we provide a comprehensive overview of the vision generalist models, delving into their characteristics and capabilities within the field. First, we review the background, including the datasets, tasks, and benchmarks. Then, we dig into the design of frameworks that have been proposed in existing research, while also introducing the techniques employed to enhance their performance. To better help the researchers comprehend the area, we take a brief excursion into related domains, shedding light on their interconnections and potential synergies. To conclude, we provide some real-world application scenarios, undertake a thorough examination of the pe
Vision graph neural networks (ViG) offer a new avenue for exploration in computer vision. A major bottleneck in ViGs is the inefficient k-nearest neighbor (KNN) operation used for graph construction. To solve this issue, we propose a new method for designing ViGs, Dynamic Axial Graph Construction (DAGC), which is more efficient than KNN as it limits the number of considered graph connections made within an image. Additionally, we propose a novel CNN-GNN architecture, GreedyViG, which uses DAGC. Extensive experiments show that GreedyViG beats existing ViG, CNN, and ViT architectures in terms of accuracy, GMACs, and parameters on image classification, object detection, instance segmentation, and semantic segmentation tasks. Our smallest model, GreedyViG-S, achieves 81.1% top-1 accuracy on ImageNet-1K, 2.9% higher than Vision GNN and 2.2% higher than Vision HyperGraph Neural Network (ViHGNN), with less GMACs and a similar number of parameters. Our largest model, GreedyViG-B obtains 83.9% top-1 accuracy, 0.2% higher than Vision GNN, with a 66.6% decrease in parameters and a 69% decrease in GMACs. GreedyViG-B also obtains the same accuracy as ViHGNN with a 67.3% decrease in parameters a
The systematic evaluation and understanding of computer vision models under varying conditions require large amounts of data with comprehensive and customized labels, which real-world vision datasets rarely satisfy. While current synthetic data generators offer a promising alternative, particularly for embodied AI tasks, they often fall short for computer vision tasks due to low asset and rendering quality, limited diversity, and unrealistic physical properties. We introduce the BEHAVIOR Vision Suite (BVS), a set of tools and assets to generate fully customized synthetic data for systematic evaluation of computer vision models, based on the newly developed embodied AI benchmark, BEHAVIOR-1K. BVS supports a large number of adjustable parameters at the scene level (e.g., lighting, object placement), the object level (e.g., joint configuration, attributes such as "filled" and "folded"), and the camera level (e.g., field of view, focal length). Researchers can arbitrarily vary these parameters during data generation to perform controlled experiments. We showcase three example application scenarios: systematically evaluating the robustness of models across different continuous axes of d
In recent years, there has been an explosion of 2D vision models for numerous tasks such as semantic segmentation, style transfer or scene editing, enabled by large-scale 2D image datasets. At the same time, there has been renewed interest in 3D scene representations such as neural radiance fields from multi-view images. However, the availability of 3D or multiview data is still substantially limited compared to 2D image datasets, making extending 2D vision models to 3D data highly desirable but also very challenging. Indeed, extending a single 2D vision operator like scene editing to 3D typically requires a highly creative method specialized to that task and often requires per-scene optimization. In this paper, we ask the question of whether any 2D vision model can be lifted to make 3D consistent predictions. We answer this question in the affirmative; our new Lift3D method trains to predict unseen views on feature spaces generated by a few visual models (i.e. DINO and CLIP), but then generalizes to novel vision operators and tasks, such as style transfer, super-resolution, open vocabulary segmentation and image colorization; for some of these tasks, there is no comparable previou
Finding the minimum spanning tree (MST) of a graph is an important task in computer vision, as it enables a sparse and low-cost representation of connectivity among elements (such as superpixels, points, or regions), which is useful for tasks such as segmentation, reconstruction, and clustering. In this work, we propose and evaluate a fully quantum pipeline for computing MSTs using the FALQON algorithm, a feedback-based quantum optimization method that does not require classical optimizers. We construct a Hamiltonian formulation whose ground-state energy encodes the MST of a graph and compare different FALQON strategies: (i) time rescaling (TR-FALQON) and (ii) multi-driver configurations. To avoid domain-specific biases, we adopt graphs with random weights and show that the FALQON variants exhibit significant differences in ground-state fidelity. We discuss the relevance of this approach for computer vision problems that naturally yield graph representations, and experimental results on synthetic instances together with a small demonstrative study on image segmentation illustrate both the potential and the current limitations of the method. Our numerical simulations on randomly wei
Reasoning about fine-grained spatial relationships in warehouse-scale environments poses a significant challenge for existing vision-language models (VLMs), which often struggle to comprehend 3D layouts, object arrangements, and multimodal cues in real-world industrial settings. In this paper, we present TinyGiantVLM, a lightweight and modular two-stage framework designed for physical spatial reasoning, distinguishing itself from traditional geographic reasoning in complex logistics scenes. Our approach encodes both global and region-level features from RGB and depth modalities using pretrained visual backbones. To effectively handle the complexity of high-modality inputs and diverse question types, we incorporate a Mixture-of-Experts (MoE) fusion module, which dynamically combines spatial representations to support downstream reasoning tasks and improve convergence. Training is conducted in a two-phase strategy: the first phase focuses on generating free-form answers to enhance spatial reasoning ability, while the second phase uses normalized answers for evaluation. Evaluated on Track 3 of the AI City Challenge 2025, our 64M-parameter base model achieved 5th place on the leaderboa
Eye contact is a crucial non-verbal interaction modality and plays an important role in our everyday social life. While humans are very sensitive to eye contact, the capabilities of machines to capture a person's gaze are still mediocre. We tackle this challenge and present NITEC, a hand-annotated eye contact dataset for ego-vision interaction. NITEC exceeds existing datasets for ego-vision eye contact in size and variety of demographics, social contexts, and lighting conditions, making it a valuable resource for advancing ego-vision-based eye contact research. Our extensive evaluations on NITEC demonstrate strong cross-dataset performance, emphasizing its effectiveness and adaptability in various scenarios, that allows seamless utilization to the fields of computer vision, human-computer interaction, and social robotics. We make our NITEC dataset publicly available to foster reproducibility and further exploration in the field of ego-vision interaction. https://github.com/thohemp/nitec
Transformers are widely used as generic backbones in computer vision, despite initially introduced for natural language processing. Recently, the Long Short-Term Memory (LSTM) has been extended to a scalable and performant architecture - the xLSTM - which overcomes long-standing LSTM limitations via exponential gating and parallelizable matrix memory structure. In this report, we introduce Vision-LSTM (ViL), an adaption of the xLSTM building blocks to computer vision. ViL comprises a stack of xLSTM blocks where odd blocks process the sequence of patch tokens from top to bottom while even blocks go from bottom to top. Experiments show that ViL holds promise to be further deployed as new generic backbone for computer vision architectures.
Facial video-based remote physiological measurement is a promising research area for detecting human vital signs (e.g., heart rate, respiration frequency) in a non-contact way. Conventional approaches are mostly supervised learning, requiring extensive collections of facial videos and synchronously recorded photoplethysmography (PPG) signals. To tackle it, self-supervised learning has recently gained attentions; due to the lack of ground truth PPG signals, its performance is however limited. In this paper, we propose a novel self-supervised framework that successfully integrates the popular vision-language models (VLMs) into the remote physiological measurement task. Given a facial video, we first augment its positive and negative video samples with varying rPPG signal frequencies. Next, we introduce a frequency-oriented vision-text pair generation method by carefully creating contrastive spatio-temporal maps from positive and negative samples and designing proper text prompts to describe their relative ratios of signal frequencies. A pre-trained VLM is employed to extract features for these formed vision-text pairs and estimate rPPG signals thereafter. We develop a series of gener
Vision transformers have shown great success on numerous computer vision tasks. However, its central component, softmax attention, prohibits vision transformers from scaling up to high-resolution images, due to both the computational complexity and memory footprint being quadratic. Although linear attention was introduced in natural language processing (NLP) tasks to mitigate a similar issue, directly applying existing linear attention to vision transformers may not lead to satisfactory results. We investigate this problem and find that computer vision tasks focus more on local information compared with NLP tasks. Based on this observation, we present a Vicinity Attention that introduces a locality bias to vision transformers with linear complexity. Specifically, for each image patch, we adjust its attention weight based on its 2D Manhattan distance measured by its neighbouring patches. In this case, the neighbouring patches will receive stronger attention than far-away patches. Moreover, since our Vicinity Attention requires the token length to be much larger than the feature dimension to show its efficiency advantages, we further propose a new Vicinity Vision Transformer (VVT) st
Prompt learning has emerged as an efficient and effective method for fine-tuning vision-language models such as CLIP. While many studies have explored generalisation abilities of these models in few-shot classification tasks and a few studies have addressed far out-of-distribution (OOD) of the models, their potential for addressing near OOD detection remains underexplored. Existing methods either require training from scratch, need fine-tuning, or are not designed for vision-language prompt learning. To address this, we introduce the Contrastive Logit Score (CLS), a novel post-hoc, plug-and-play scoring function. CLS significantly improves near OOD detection of pre-trained vision-language prompt learning methods without modifying their model architectures or requiring retraining. Our method achieves up to an 11.67% improvement in AUROC for near OOD detection with minimal computational overhead. Extensive evaluations validate the effectiveness, efficiency, and generalisability of our approach. Our code is available at https://github.com/davidmcjung/near-OOD-prompt-learning.
This paper presents a novel approach that integrates vision foundation models with reinforcement learning to enhance object interaction capabilities in simulated environments. By combining the Segment Anything Model (SAM) and YOLOv5 with a Proximal Policy Optimization (PPO) agent operating in the AI2-THOR simulation environment, we enable the agent to perceive and interact with objects more effectively. Our comprehensive experiments, conducted across four diverse indoor kitchen settings, demonstrate significant improvements in object interaction success rates and navigation efficiency compared to a baseline agent without advanced perception. The results show a 68% increase in average cumulative reward, a 52.5% improvement in object interaction success rate, and a 33% increase in navigation efficiency. These findings highlight the potential of integrating foundation models with reinforcement learning for complex robotic tasks, paving the way for more sophisticated and capable autonomous agents.