Zilun Zhang

Zilun Zhang, Tiancheng Zhao, Yulong Guo et al.

Pre-trained Vision-Language Models (VLMs) utilizing extensive image-text paired data have demonstrated unprecedented image-text association capabilities, achieving remarkable results across various downstream tasks. A critical challenge is how to make use of existing large-scale pre-trained VLMs, which are trained on common objects, to perform the domain-specific transfer for accomplishing domain-related downstream tasks. A critical challenge is how to make use of existing large-scale pre-trained VLMs, which are trained on common objects, to perform the domain-specific transfer for accomplishing domain-related downstream tasks. In this paper, we propose a new framework that includes the Domain pre-trained Vision-Language Model (DVLM), bridging the gap between the General Vision-Language Model (GVLM) and domain-specific downstream tasks. Moreover, we present an image-text paired dataset in the field of remote sensing (RS), RS5M, which has 5 million RS images with English descriptions. The dataset is obtained from filtering publicly available image-text paired datasets and captioning label-only RS datasets with pre-trained VLM. These constitute the first large-scale RS image-text paired dataset. Additionally, we fine-tuned the CLIP model and tried several Parameter-Efficient Fine-Tuning methods on RS5M to implement the DVLM. Experimental results show that our proposed dataset is highly effective for various tasks, and our model GeoRSCLIP improves upon the baseline or previous state-of-the-art model by $3\%\sim20\%$ in Zero-shot Classification (ZSC), $3\%\sim6\%$ in Remote Sensing Cross-Modal Text-Image Retrieval (RSCTIR) and $4\%\sim5\%$ in Semantic Localization (SeLo) tasks. Dataset and models have been released in: \url{https://github.com/om-ai-lab/RS5M}.

Zilun Zhang, Cuifeng Shen, Yuan Shen et al.

Although CLIP-like Visual Language Models provide a functional joint feature space for image and text, due to the limitation of the CILP-like model's image input size (e.g., 224), subtle details are lost in the feature representation if we input high-resolution images (e.g., 2240). Our proposed framework addresses this issue by generating a single feature representation for a high-resolution image that retains image details from different scales while sharing the same semantic space as the original CLIP. An application scenario is remote sensing text-image retrieval, where targets (e.g., vehicles and ships) often appear at tiny scales. To achieve this, we develop a feature fusion model that relies on CLIP features extracted from a carefully designed image patch method, dubbed Complete Cover. This method ensures comprehensive coverage of objects across various scales and is weakly supervised by image-agnostic class prompted queries. We evaluate our framework's performance using real-world and synthetic datasets, demonstrating significant improvements in image retrieval tasks based on class prompted queries. To further showcase our framework's capability in detail retrieval, we introduce a CLEVR-like synthetic dataset, named CLVER-DS. This fully annotated dataset offers a controllable object scale, allowing for a more thorough examination of our approach's effectiveness.Our code is publicly available at https://github.com/zilunzhang/DetailCLIP

Zilun Zhang, Farzad Khalvati

Many high-performance classification models utilize complex CNN-based architectures for Alzheimer's Disease classification. We aim to investigate two relevant questions regarding classification of Alzheimer's Disease using MRI: "Do Vision Transformer-based models perform better than CNN-based models?" and "Is it possible to use a shallow 3D CNN-based model to obtain satisfying results?" To achieve these goals, we propose two models that can take in and process 3D MRI scans: Convolutional Voxel Vision Transformer (CVVT) architecture, and ConvNet3D-4, a shallow 4-block 3D CNN-based model. Our results indicate that the shallow 3D CNN-based models are sufficient to achieve good classification results for Alzheimer's Disease using MRI scans.

Haozhan Shen, Peng Liu, Jingcheng Li et al. · cmu

Recently DeepSeek R1 has shown that reinforcement learning (RL) can substantially improve the reasoning capabilities of Large Language Models (LLMs) through a simple yet effective design. The core of R1 lies in its rule-based reward formulation, which leverages tasks with deterministic ground-truth answers to enable precise and stable reward computation. In the visual domain, we similarly observe that a wide range of visual understanding tasks are inherently equipped with well-defined ground-truth annotations. This property makes them naturally compatible with rule-based reward mechanisms. Motivated by this observation, we investigate the extension of R1-style reinforcement learning to Vision-Language Models (VLMs), aiming to enhance their visual reasoning capabilities. To this end, we develop VLM-R1, a dedicated framework designed to harness RL for improving VLMs' performance on general vision-language tasks. Using this framework, we further explore the feasibility of applying RL to visual domain. Experimental results indicate that the RL-based model not only delivers competitive performance on visual understanding tasks but also surpasses Supervised Fine-Tuning (SFT) in generalization ability. Furthermore, we conduct comprehensive ablation studies that uncover a series of noteworthy insights, including the presence of reward hacking in object detection, the emergence of the "OD aha moment", the impact of training data quality, and the scaling behavior of RL across different model sizes. Through these analyses, we aim to deepen the understanding of how reinforcement learning enhances the capabilities of vision-language models, and we hope our findings and open-source contributions will support continued progress in the vision-language RL community. Our code and model are available at https://github.com/om-ai-lab/VLM-R1

Yue Zhou, Jue Chen, Zilun Zhang et al.

Remote sensing (RS) large vision-language models (LVLMs) have shown strong promise across visual grounding (VG) tasks. However, existing RS VG datasets predominantly rely on explicit referring expressions-such as relative position, relative size, and color cues-thereby constraining performance on implicit VG tasks that require scenario-specific domain knowledge. This article introduces DVGBench, a high-quality implicit VG benchmark for drones, covering six major application scenarios: traffic, disaster, security, sport, social activity, and productive activity. Each object provides both explicit and implicit queries. Based on the dataset, we design DroneVG-R1, an LVLM that integrates the novel Implicit-to-Explicit Chain-of-Thought (I2E-CoT) within a reinforcement learning paradigm. This enables the model to take advantage of scene-specific expertise, converting implicit references into explicit ones and thus reducing grounding difficulty. Finally, an evaluation of mainstream models on both explicit and implicit VG tasks reveals substantial limitations in their reasoning capabilities. These findings provide actionable insights for advancing the reasoning capacity of LVLMs for drone-based agents. The code and datasets will be released at https://github.com/zytx121/DVGBench

Haozhan Shen, Kangjia Zhao, Tiancheng Zhao et al.

Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities in vision-language understanding. Recently, with the integration of test-time scaling techniques, these models have also shown strong potential in visual reasoning. However, most existing reasoning approaches remain text-level in nature: MLLMs are prompted to explore various combinations of textual tokens via their underlying language model, while the visual input remains fixed throughout the reasoning process. This paradigm limits the model's ability to fully exploit rich visual information, particularly when dealing with images containing numerous fine-grained elements. In such cases, vision-level reasoning becomes crucial - where models dynamically zoom into specific regions of the image to gather detailed visual cues necessary for accurate decision-making. In this paper, we propose Zoom Eye, a training-free, model-agnostic tree search algorithm tailored for vision-level reasoning. Zoom Eye treats an image as a hierarchical tree structure, where each child node represents a zoomed-in sub-region of its parent, and the root corresponds to the full image. The algorithm enables MLLMs to simulate human-like zooming behavior by navigating from root to leaf nodes in search of task-relevant visual evidence. We experiment on a series of high-resolution benchmarks and the results demonstrate that Zoom Eye consistently improves the performance of multiple MLLMs by a large margin (e.g., InternVL2.5-8B increases by 15.71% and 17.69% on HR-Bench) and also enables small 3-8B MLLMs to outperform strong large models such as GPT-4o. Code: https://github.com/om-ai-lab/ZoomEye

Zilun Zhang, Cuifeng Shen, Yuan Shen et al.

Although CLIP-like Visual Language Models provide a functional joint feature space for image and text, due to the limitation of the CILP-like model's image input size (e.g., 224), subtle details are lost in the feature representation if we input high-resolution images (e.g., 2240). In this work, we introduce an efficient framework that can produce a single feature representation for a high-resolution image that injects image details and shares the same semantic space as the original CLIP. In the framework, we train a feature fusing model based on CLIP features extracted from a carefully designed image patch method that can cover objects of any scale, weakly supervised by image-agnostic class prompted queries. We validate our framework by retrieving images from class prompted queries on the real world and synthetic datasets, showing significant performance improvement on these tasks. Furthermore, to fully demonstrate our framework's detail retrieval ability, we construct a CLEVR-like synthetic dataset called CLVER-DS, which is fully annotated and has a controllable object scale.

Zilun Zhang, Haozhan Shen, Tiancheng Zhao et al. · cmu

Ultra High Resolution (UHR) remote sensing imagery (RSI) (e.g. 100,000 $\times$ 100,000 pixels or more) poses a significant challenge for current Remote Sensing Multimodal Large Language Models (RSMLLMs). If choose to resize the UHR image to standard input image size, the extensive spatial and contextual information that UHR images contain will be neglected. Otherwise, the original size of these images often exceeds the token limits of standard RSMLLMs, making it difficult to process the entire image and capture long-range dependencies to answer the query based on the abundant visual context. In this paper, we introduce ImageRAG for RS, a training-free framework to address the complexities of analyzing UHR remote sensing imagery. By transforming UHR remote sensing image analysis task to image's long context selection task, we design an innovative image contextual retrieval mechanism based on the Retrieval-Augmented Generation (RAG) technique, denoted as ImageRAG. ImageRAG's core innovation lies in its ability to selectively retrieve and focus on the most relevant portions of the UHR image as visual contexts that pertain to a given query. Fast path and slow path are proposed in this framework to handle this task efficiently and effectively. ImageRAG allows RSMLLMs to manage extensive context and spatial information from UHR RSI, ensuring the analysis is both accurate and efficient. Codebase will be released in https://github.com/om-ai-lab/ImageRAG

Yutao Sun, Mingshuai Chen, Tiancheng Zhao et al.

Catastrophic forgetting remains a major challenge when fine-tuning large language models (LLMs) on narrow, task-specific data, often degrading their general knowledge and reasoning abilities. We propose SA-SFT, a lightweight self-augmentation routine in which an LLM generates self-dialogues prior to fine-tuning, and the resulting self-authored data are mixed with task data without modifying optimization or training schedules. Despite requiring no external data or additional tuning, SA-SFT consistently mitigates catastrophic forgetting while improving in-domain performance. Across 50 evaluation scenarios, it maintains performance comparable to the original model and achieves the best results in 40 cases, outperforming common baselines such as layer freezing and external data mixing. Guided by these empirical findings, we further present a theoretical analysis suggesting that forgetting can partly stem from style-induced parameter drift, and that self-alignment through self-generated data provides an effective means to counteract this effect. Overall, our results indicate that self-augmentation offers a simple and effective mechanism for robust LLM adaptation without incurring catastrophic forgetting.

Zilun Zhang, Zian Guan, Tiancheng Zhao et al.

Referring expression understanding in remote sensing poses unique challenges, as it requires reasoning over complex object-context relationships. While supervised fine-tuning (SFT) on multimodal large language models achieves strong performance with massive labeled datasets, they struggle in data-scarce scenarios, leading to poor generalization. To address this limitation, we propose Geo-R1, a reasoning-centric reinforcement fine-tuning (RFT) paradigm for few-shot geospatial referring. Geo-R1 enforces the model to first generate explicit, interpretable reasoning chains that decompose referring expressions, and then leverage these rationales to localize target objects. This "reason first, then act" process enables the model to make more effective use of limited annotations, enhances generalization, and provides interpretability. We validate Geo-R1 on three carefully designed few-shot geospatial referring benchmarks, where our model consistently and substantially outperforms SFT baselines. It also demonstrates strong cross-dataset generalization, highlighting its robustness. Code and data will be released at: https://github.com/Geo-R1/geo-r1.

Yulong Guo, Zilun Zhang, Yongheng Shang et al.

The long-tail problem presents a significant challenge to the advancement of semantic segmentation in ultra-high-resolution (UHR) satellite imagery. While previous efforts in UHR semantic segmentation have largely focused on multi-branch network architectures that emphasize multi-scale feature extraction and fusion, they have often overlooked the importance of addressing the long-tail issue. In contrast to prior UHR methods that focused on independent feature extraction, we emphasize data augmentation and multimodal feature fusion to alleviate the long-tail problem. In this paper, we introduce SRMF, a novel framework for semantic segmentation in UHR satellite imagery. Our approach addresses the long-tail class distribution by incorporating a multi-scale cropping technique alongside a data augmentation strategy based on semantic reordering and resampling. To further enhance model performance, we propose a multimodal fusion-based general representation knowledge injection method, which, for the first time, fuses text and visual features without the need for individual region text descriptions, extracting more robust features. Extensive experiments on the URUR, GID, and FBP datasets demonstrate that our method improves mIoU by 3.33\%, 0.66\%, and 0.98\%, respectively, achieving state-of-the-art performance. Code is available at: https://github.com/BinSpa/SRMF.git.

Yutao Sun, Mingshuai Chen, Tiancheng Zhao et al. · cmu

Self-improving large language models (LLMs) -- i.e., to improve the performance of an LLM by fine-tuning it with synthetic data generated by itself -- is a promising way to advance the capabilities of LLMs while avoiding extensive supervision. Existing approaches to self-improvement often rely on external supervision signals in the form of seed data and/or assistance from third-party models. This paper presents Crescent -- a simple yet effective framework for generating high-quality synthetic question-answer data in a fully autonomous manner. Crescent first elicits the LLM to generate raw questions via a bait prompt, then diversifies these questions leveraging a rejection sampling-based self-deduplication, and finally feeds the questions to the LLM and collects the corresponding answers by means of majority voting. We show that Crescent sheds light on the potential of true self-improvement with zero external supervision signals for math reasoning; in particular, Crescent-generated question-answer pairs suffice to (i) improve the reasoning capabilities of an LLM while preserving its general performance (especially in the 0-shot setting); and (ii) distil LLM knowledge to weaker models more effectively than existing methods based on seed-dataset augmentation.

Zilun Zhang, Haozhan Shen, Tiancheng Zhao et al.

The application of Vision-Language Models (VLMs) in remote sensing (RS) has demonstrated significant potential in traditional tasks such as scene classification, object detection, and image captioning. However, current models, which excel in Referring Expression Comprehension (REC), struggle with tasks involving complex instructions (e.g., exists multiple conditions) or pixel-level operations like segmentation and change detection. In this white paper, we provide a comprehensive hierarchical summary of vision-language tasks in RS, categorized by the varying levels of cognitive capability required. We introduce the Remote Sensing Vision-Language Task Set (RSVLTS), which includes Open-Vocabulary Tasks (OVT), Referring Expression Tasks (RET), and Described Object Tasks (DOT) with increased difficulty, and Visual Question Answering (VQA) aloneside. Moreover, we propose a novel unified data representation using a set-of-points approach for RSVLTS, along with a condition parser and a self-augmentation strategy based on cyclic referring. These features are integrated into the GeoRSMLLM model, and this enhanced model is designed to handle a broad range of tasks of RSVLTS, paving the way for a more generalized solution for vision-language tasks in geoscience and remote sensing.

Zilun Zhang, Yutao Sun, Tiancheng Zhao et al.

Humans can retain old knowledge while learning new information, but Large Language Models (LLMs) often suffer from catastrophic forgetting when post-pretrained or supervised fine-tuned (SFT) on domain-specific data. Moreover, for Multimodal Large Language Models (MLLMs) which are composed of the LLM base and visual projector (e.g. LLaVA), a significant decline in performance on language benchmarks was observed compared to their single-modality counterparts. To address these challenges, we introduce a novel model-agnostic self-decompression method, Tree Generation (TG), that decompresses knowledge within LLMs into the training corpus. This paper focuses on TG-SFT, which can synthetically generate SFT data for the instruction tuning steps. By incorporating the dumped corpus during SFT for MLLMs, we significantly reduce the forgetting problem.

Zilun Zhang, Shihao Ma, Yichun Zhang

Most few-shot learning models utilize only one modality of data. We would like to investigate qualitatively and quantitatively how much will the model improve if we add an extra modality (i.e. text description of the image), and how it affects the learning procedure. To achieve this goal, we propose four types of fusion method to combine the image feature and text feature. To verify the effectiveness of improvement, we test the fusion methods with two classical few-shot learning models - ProtoNet and MAML, with image feature extractors such as ConvNet and ResNet12. The attention-based fusion method works best, which improves the classification accuracy by a large margin around 30% comparing to the baseline result.

Ling Yang, Liangliang Li, Zilun Zhang et al.

Most graph-network-based meta-learning approaches model instance-level relation of examples. We extend this idea further to explicitly model the distribution-level relation of one example to all other examples in a 1-vs-N manner. We propose a novel approach named distribution propagation graph network (DPGN) for few-shot learning. It conveys both the distribution-level relations and instance-level relations in each few-shot learning task. To combine the distribution-level relations and instance-level relations for all examples, we construct a dual complete graph network which consists of a point graph and a distribution graph with each node standing for an example. Equipped with dual graph architecture, DPGN propagates label information from labeled examples to unlabeled examples within several update generations. In extensive experiments on few-shot learning benchmarks, DPGN outperforms state-of-the-art results by a large margin in 5% $\sim$ 12% under supervised setting and 7% $\sim$ 13% under semi-supervised setting. Code will be released.