Aug 2026· IEEE Geoscience and Remote Sensing Letters· Vol 23, pp. 4014105-4014105· 0 citations· 16 references
Computer Science
TL;DR
Results on the UC Merced (UCM) and NWPU benchmarks indicate that SE-CLIP significantly outperforms existing semi-supervised approaches and provides a viable solution for adapting VLMs to the remote sensing domain with minimal human intervention.
Abstract
Vision-language models (VLMs), like contrastive language-image pretraining (CLIP), have shown significant potential in handling natural images, yet their performance is often limited by the distinct characteristics of satellite imagery. While parameter-efficient adaptation techniques exist, their efficacy is frequently limited by the scarcity of annotated samples. In this letter, we propose self-evolutionary CLIP (SE-CLIP), a semisupervised framework designed for recursive label mining in scene classification. The approach follows a dual-phase pipeline, where an initial warmup on a few annotated seeds is followed by a recursive discovery phase that iteratively identifies high-confidence samples from unlabeled pools. To maintain the integrity of the evolving support set, we employ a class-balanced selection strategy that prevents the model from being dominated by easily learned categories. Results on the UC Merced (UCM) and NWPU benchmarks indicate that SE-CLIP significantly outperforms existing semi-supervised approaches. The framework provides a viable solution for adapting VLMs to the remote sensing (RS) domain with minimal human intervention.
Across cross-generator, post-processing, and in-the-wild benchmarks, PE-SPC surpasses the previous DINOv3 baseline and achieves new state-of-the-art results.
Wei-Han Cai, Hao Tan, Zichang Tan et al.· 0 citations
Existing class-incremental learning (CIL) methods for remote sensing (RS) scene classification often tend to be training-intensive or rely on static visual features that may inadequately capture the complex interclass similarity and intraclass diversity inherent in RS imagery. Moreover, directly reusing features from models pretrained on natural images (e.g., CLIP) often suffers from a domain gap and limited semantic richness when transferred to the RS domain. In this article, we propose a simple yet effective training-free CIL framework for RS scene classification that leverages multimodal semantic information to build more discriminative category representations. Our framework treats pretrained models as frozen feature extractors to guarantee zero forgetting of the representation space. To enhance semantic discriminability, we employ a large-language model (LLM) to generate rich candidate textual descriptions for each class and introduce an image-guided description selection (IGDS) strategy to align semantic information with visual characteristics. Classification is performed using a distance-based metric without any additional training. Extensive experimental results demonstrate that our framework achieves leading performance and superior stability across different session sequences, surpassing both training-based and training-free baseline methods. Our code is available at https://github.com/WenliangDu/ZFCIL-RS
Wen-Liang Du, Ji-Cun He, Jiaqi Zhao et al.· IEEE Transactions on Geoscie...· 0 citations
Compared to prior CLIP-enhancement methods, MLLMCLIP achieves state-of-the-art compositional accuracy while delivering consistent gains on standard zero-shot classification and image-text retrieval, showing that feature-level distillation strengthens both compositional and general vision-language representation capability.
Jongsuk Kim, Qiyu Wu, Zhuoyuan Mao et al.· 0 citations
UC-VLM is a unified multi-stage binary-supervised framework that consistently reuses the same authenticity labels for visual adaptation and label-conditioned text generation, while leveraging automatically optimized instructions to reduce prompt sensitivity without requiring human-written rationales or hand-crafted prompts.
Lei Tan, Shuwei Li, Mohan S. Kankanhalli et al.· 0 citations
Recent Visual-Language Models (VLMs) have enhanced the capabilities of pre-trained LLMs by adding vision tokens alongside text, with approaches like LLaVA showing impressive results. However, the computational burden of processing up to 576 or 729 visual tokens makes edge deployment challenging. While various token pruning techniques require retraining, some are training-free and thus can easily adapt to architecture changes. We introduce ClustRS, a two-part, training-free algorithm for robust token pruning. Its first component is an attention-weighted, clustering algorithm that selects representative tokens from each semantic cluster. The second component, Residual Shrinkage, is a one-pass denoising step on the selected tokens. These training-free lightweight steps make LLaVA ready for real-world data, improving robustness to a wide range of image-noise types and intensities. Experimental results on the ScienceQA-IMG and MM-VET benchmarks show our method outperforms attention- and diversity-based methods by up to 20\% under extreme noise and token conditions (reducing tokens by 97\%, down to 16 tokens) on LLaVA 1.5 7b and achieves exceptional results on LLaVA-OneVision, where we match baseline performance with fewer than one-third of their tokens under mild noise conditions. Our study demonstrates a simple yet powerful alternative to both score-only and diversity-only pruning rules, paving the way for compute-efficient and noise-resilient VLM deployment.
Baptiste Rossigneux, Inna Kucher, Vincent Lorrain et al.· 0 citations
