Penalty kicks in soccer are decided under extreme time constraints, where goalkeepers benefit from anticipating shot direction from the kickers motion before or around ball contact. In this paper, MambaKick is presented as a learning-based framework for penalty direction prediction that leverages pretrained human action recognition (HAR) embeddings extracted from contact-centered short video segments and combines them with a lightweight temporal predictor. Rather than relying on explicit kinematic reconstruction or handcrafted biomechanical features, the approach reuses transferable spatiotemporal representations and utilizes selective state-spare models (Mamba) for efficient sequence aggregation. Simple contextual metadata (e.g., field side and footedness) are also considered as complementary cues that may reduce ambiguity in real-world footage. Across a range of HAR backbones, MambaKick consistently improves or matches strong embedding baselines, achieving up to 53.1

This paper presents a novel method to remove non-uniform haze in real images. The process involves complex phases: initial extraction of image features, haze removal, and final image reconstruction. To address this complex challenge, the proposed approach implements a transformer-based architecture. By extending the SwinIR image restoration framework, the method includes modifications to the deep feature extraction module and also implements adaptive tokenization and learnable position embeddings. The results show significant advances with respect to existing models, validating the effectiveness of the proposed strategy to remove non-homogeneous haze within the images.

In this study, we explore an enhancement to the U-Net architecture by integrating SK-ResNeXt as the encoder for Land Cover Classification (LCC) tasks using Multispectral Imaging (MSI). SK-ResNeXt introduces cardinality and adaptive kernel sizes, allowing U-Net to better capture multi-scale features and adjust more effectively to variations in spatial resolution, thereby enhancing the model’s ability to segment complex land cover types. We evaluate this approach using the Five-Billion-Pixels dataset, composed of 150 large-scale RGB-NIR images and over 5 billion labeled pixels across 24 categories. The approach achieves notable improvements over the baseline U-Net, with gains of 5.312% in Overall Accuracy (OA) and 8.906% in mean Intersection over Union (mIoU) when using the RGB configuration. With the RG-NIR configuration, these improvements increase to 6.928% in OA and 6.938% in mIoU, while the RGB-NIR configuration yields gains of 5.854% in OA and 7.794% in mIoU. Furthermore, the approach not only outperforms other well-established models such as DeepLabV3, DeepLabV3+, Ma-Net, SegFormer, and PSPNet, particularly with the RGB-NIR configuration, but also surpasses recent state-of-the-art methods. Visual tests confirmed this superiority, showing that the studied approach achieves notable improvements in certain classes, such as lakes, rivers, industrial areas, residential areas, and vegetation, where the other architectures struggled to achieve accurate segmentation. These results demonstrate the potential and capability of the explored approach to effectively handle MSI and enhance LCC results.

This paper presents the results of the Sixth Thermal Image Super-Resolution Challenge held within the Perception Beyond the Visible Spectrum (PBVS) workshop at CVPR 2025. The challenge maintains the same cross-spectral benchmark dataset as the previous year, consisting of 1000 thermal images each paired with corresponding high-resolution RGB images. Track 1 focuses on single thermal image super-resolution, enhancing low-resolution infrared images by a factor of x 8, whereas Track 2 addresses guided thermal image super-resolution, performing super-resolution at scale factors of x 8 and x 16 by leveraging high-resolution RGB images as auxiliary inputs. The 2025 edition attracted increased participation, with 128 teams competing in Track 1 and 86 teams in Track 2. The paper describes methodologies employed by the top participating teams, emphasizing innovations in transformer-based and hybrid architectures, and provides a detailed comparative analysis of the results between the 2024 and 2025 challenges. This analysis reveals significant progress in thermal image reconstruction accuracy, showcasing notable advances achieved by the leading methodologies.

This paper proposes a novel approach for obtaining high-quality thermal image-like representations that can be used as inputs in various thermal image compressive sensing applications. We address the challenge of low-resolution/quality thermal images by generating synthetic thermal image representations using a contrastive cycled GAN network from low-cost visible images. These representations can then be used to improve the quality of low-quality thermal images of the same scene. Experimental results demonstrate the effectiveness of the proposed approach on different datasets.

This work presents an encoder-decoder architecture designed for Land Cover Classification utilizing Multispectral Imagery. The architecture consists of two parallel feature extraction branches based on ConvNeXt, one dedicated to processing RGB information and the other to NIR data. The encoders are connected with a spectral fusion and pyramidal decoding process enhanced with a convolutional block attention module, which emphasizes the most relevant features to produce accurate segmentation masks. Experiments are conducted using the Potsdam dataset, and results demonstrate that the proposed approach achieves notable performance, with an overall accuracy of 90.83% and a mean intersection over union of 77.69%. Furthermore, visual tests show that the proposed approach is better at differentiating complex classes, achieving segmentations closer to the ground-truth. The method outperforms other well-known architectures, including U-Net, PSPNet, DeepLabV3, and DeepLabV3+, as well as other state-of-the-art approaches. Code is available at https://github.com/Leo-Thomas/mecafnet.git

This paper highlights the objectives, metrics and top performers in the 3rd Multi/Cross Modal Aerial Imagery Translation Challenge (MAVIC-T) of the 21st CVPR PBVS workshop. The core goal of this competition remains the seeding of innovative model development for translating registered aerial images between diverse sensor modalities. Specifically, the challenge explores the transformation between synthetic aperture radar (SAR), electrooptical (EO), visible light (RGB), and infrared (IR) imagery in challenging real world conditions. The competition once again judges its entrants using a composite of the L1-norm, Learned Perceptual Image Patch Similarity (LPIPS), and the Fréchet Inception Distance (FID). An additional penalty for overfitting to one domain ups the challenge from 2024, while pushing for more generalizable solutions on the second year return of the Multi Modal Aerial Gathered Image Composite Stacks (MAGIC-STACKS) Dataset. Overall, this year saw 103 total participants. The top performers scored comparably overall to last year's winners, however, there was a notable improvement this year in the RGB→IR translation task. Interestingly, the winning team - up6 was not the best across all translation scenarios, signaling room for improvement in coming years.

Tomato leaf disease detection is critical in precision agriculture for safeguarding crop health and optimizing yields. This study compares the latest YOLO architectures, including YOLOv8, YOLOv9, YOLOv10, YOLOv11, and YOLOv12, using the Tomato-Village dataset, which contains 14,368 images across six disease classes. All models are trained under identical settings to ensure a fair evaluation based on precision, recall, mean Average Precision, training time, and inference speed. Results show that YOLOv11 consistently outperforms the other architectures, achieving the highest accuracy with competitive training times and acceptable latency. YOLOv10, YOLOv8, and YOLOv12 also deliver strong results, with YOLOv12n emerging as the most effective lightweight model for resource-constrained environments. In contrast, YOLOv9 demonstrates the weakest performance, requiring more training time and exhibiting higher latency. Overall, YOLOv11 is positioned as the most effective solution for tomato leaf disease detection, providing a strong benchmark for future advancements in agricultural technology.

Multiple Object Tracking (MOT) has seen significant advancements in the RGB domain, yet remains underexplored in thermal imaging, despite its advantages in low-light and adverse weather conditions. The Thermal Pedestrian Multiple Object Tracking (TP-MOT) Challenge addresses this gap by introducing a large-scale thermal dataset and a standardized evaluation framework. This challenge provides a benchmark for tracking algorithms designed specifically for thermal data, emphasizing robust detection, motion modeling, and identity association in infrared imagery. Participants were required to use a tracking-by-detection pipeline with standardized YOLO-based detectors, ensuring a fair comparison of tracking methodologies. The topperforming approaches leveraged adaptive hyperparameter tuning, motion-based association, and infrared-specific feature extraction to enhance tracking accuracy while maintaining computational efficiency. The results demonstrate that thermal MOT can achieve high performance with dedicated methodologies, offering new insights into tracking pedestrians in challenging conditions. In this first edition a total of 11 teams have been registered for participation. This challenge serves as a catalyst for future research, paving the way for improved thermal tracking solutions in surveillance, autonomous navigation, and security applications.

This review marks the tenth anniversary of You Only Look Once (YOLO), one of the most influential frameworks in real-time object detection. Over the past decade, YOLO has evolved from a streamlined detector into a diverse family of architectures characterized by efficient design, modular scalability, and cross-domain adaptability. The paper presents a technical overview of the main versions (from YOLOv1 to YOLOv13), highlights key architectural trends, and surveys the principal application areas in which YOLO has been adopted. It also addresses evaluation practices, ethical considerations, and potential future directions for the framework’s continued development. The analysis aims to provide a comprehensive and critical perspective on YOLO’s trajectory and ongoing transformation.

The Multi-modal Aerial View Image Challenge - Classification Track (MAVIC-C) continues to push the boundaries of multi-modal object recognition by encouraging researchers to innovate models that leverage both Synthetic Aperture Radar (SAR) and Electro-Optical (EO) imagery. This paper analyzes the outcomes of the new iteration of this challenge and emphasizes the critical role of EO and SAR data fusion in remote sensing tasks. This year MAVIC-C saw impressive developments of sophisticated multi-modal approaches that address the distinct properties and challenges inherent to the data. This year's challenge notably builds on insights from previous iterates: in 2021 we demonstrated the potential of EO and SAR integration; in 2022 and 2023 we explored the capabilities of multi-modal frameworks; and in 2024 we examined model robustness in out-of-distribution scenarios. This year, we started with the same challenge design as 2024 and asked teams to further advance techniques for improving accuracy and of out-of-distribution detection, which builds model robustness. Overall, this manuscript provides an in-depth investigation of the methodologies of top-performing teams and analyzes participant's performance on a sequestered test set.

This paper proposes a novel approach to generate thermal-like representations from RGB images by using the corresponding depth map as an additional constraint. The given RGB images are converted to the HSV color space and the brightness channel is used as input together with the spatial information provided by the depth map of the given scene. This depth map is used as prior information by the generative network. By training a generative model with paired input images and their corresponding depth maps, the model learns the mapping from the RGB images to thermal-like representations. Experimental results demonstrate that the method outperforms state-of-the-art approaches, producing superior-quality thermal images with improved shape and sharpness, attributed to using depth maps as complement information.

This manuscript presents a review of state-of-the-art techniques proposed in the literature for multimodal image registration, addressing instances where images from different modalities need to be precisely aligned in the same reference system. This scenario arises when the images to be registered come from different modalities, among the visible and thermal spectral bands, 3D-RGB, or flash-no flash, or NIR-visible. The review spans different techniques from classical approaches to more modern ones based on deep learning, aiming to highlight the particularities required at each step in the registration pipeline when dealing with multimodal images. It is noteworthy that medical images are excluded from this review due to their specific characteristics, including the use of both active and passive sensors or the non-rigid nature of the body contained in the image.

Land cover classification (LCC) is a process used to categorize the earth's surface into distinct land types. This classification is vital for environmental conservation, urban planning, agricultural management, and climate change research, providing essential data for sustainable decision making. The use of multispectral imaging (MSI), which captures data beyond the visible spectrum, has emerged as one of the most utilized image modalities for addressing this task. In addition, semantic segmentation techniques play a vital role in this domain, enabling the precise delineation and labeling of land cover classes within imagery. The integration of these three concepts has given rise to an intriguing and ever-evolving research field, witnessing continuous advancements aimed at enhancing multispectral semantic segmentation (MSSS) methods for LCC. Given the dynamic nature of this field, there is a need for a thorough examination of the latest trends and advancements to understand its evolving landscape. Therefore, this article presents a review of current aspects in the field of MSSS for LCC, addressing the following key points: 1) prevalent datasets and data acquisition methods; 2) preprocessing methods for managing MSI data; 3) typical metrics and evaluation criteria used for assessing performance of methods; 4) current techniques and methodologies employed; and 5) spectral bands beyond the visible spectrum commonly utilized. Through this analysis, our objective is to provide valuable insights into the current state of MSSS for LCC, contributing to the ongoing development and understanding of this dynamic field while also providing perspectives for future research directions.

This paper outlines the advancements and results of the Fifth Thermal Image Super-Resolution challenge, hosted at the Perception Beyond the Visible Spectrum CVPR 2024 workshop. The challenge employed a novel benchmark cross-spectral dataset consisting of 1000 thermal images, each paired with its corresponding registered RGB image. The challenge featured two tracks: Track-1 focused on Single Thermal Image Super-Resolution with an x8 upscale factor, while Track-2 extended its evaluation to include both x8 and x16 scaling factors, utilizing high-resolution RGB images to guide the super-resolution process for low-resolution thermal images. The participation of over 175 teams highlights the research community's strong engagement and dedication to enhancing image resolution techniques across both single and cross-spectral methodologies. This year's challenge sets new benchmarks and provides valuable insights into future directions for research in thermal image super-resolution.

This manuscript delineates the outcomes of the fourth Multi-modal Aerial View Image Challenge - Classification (MAVIC-C). The challenge is aimed at advancing the development of recognition models that leverage Synthetic Aperture Radar (SAR) and Electro-Optical (EO) imagery. Encouraging the integration of data from these two distinct modalities, the challenge seeks to foster the creation of multi-modal approaches that complement characteristics of SAR and EO information. Building upon the precedents set in previous years, the 2021 MAVOC challenge validated the potential of integrating SAR and EO modalities. The subsequent 2022 and 2023 challenges further explored the capabilities of multi-modal frameworks. In its latest iteration, the 2024 challenge presents an enhanced UNIfied COincident Optical and Radar for recognitioN (UNICORN) dataset alongside a revised competition format, focused on the task of SAR classification. The 2024 challenge evaluates model robustness through out-of-distribution measures, alongside traditional accuracy metrics. The core of this paper is devoted to analyzing the methodologies of the top-performing entries and their performance metrics on a blind test set.