Instance-aware adaptive label assignment for 3D object detection

Tiny object detection based on dynamic scale-awareness label assignment and contextual enhancement

Moving Object Detection (MOD) is the fundamental backbone of autonomous systems, urban surveillance, and industrial robotics. This paper explores the transition from traditional background subtraction to the current "Edge-First" era dominated by YOLO26 and Real-Time Detection Transformers (RT-DETR). We analyze key innovations including NMS-free inference, temporal context modeling via Vision Transformers (ViTs), and the integration of Small-Target-Aware Label Assignment (STAL) to address long-standing challenges in dynamic environments.

In recent years, the refinement of bounding box representations has emerged as a major research focus in remote sensing. Nevertheless, mainstream detection algorithms typically ignore the disruptive impacts induced by the diverse morphologies and arbitrary orientations of high-aspect-ratio aerial objects throughout model training, thereby giving rise to several critical technical challenges: (1) Anisotropic information distribution: Target features are highly concentrated in one spatial dimension but sparse in the other, with significant feature differences across bounding box parameters, breaking the symmetry of feature distribution. (2) Missing high-quality positive samples: IoU-based assignment strategies fail to adequately capture the symmetric structural characteristics of elongated targets, resulting in incomplete coverage of critical features. (3) Loss function gradient instability: Small deviations in large-aspect-ratio bounding boxes cause drastic loss value fluctuations, as the asymmetric gradient changes hinder stable optimization directions during training. To address the challenges, we propose a Spatial Orthogonal and Boundary-Aware Network (SOBA-Net) for rotated and elongated target detection, leveraging symmetry-aware designs to enhance feature representation. Specifically, spatial staggered convolutions are constructed to fuse local and directional contextual features, effectively modeling long-range symmetric information across multiple spatial scales and reducing background noise interference. Secondly, the designed Symmetric-Constrained Label Assignment (SC-LA) introduces an IoU-weighted function, ensuring high-quality samples with symmetric structural features are classified as positive samples. Ultimately, the designed Gradient Dynamic Equilibrium Loss Function mitigates the problem of unstable gradients associated with high-aspect-ratio objects by enforcing symmetrical gradient regulation across samples with negligible localization deviations. Comprehensive evaluations across three representative remote sensing benchmarks—DOTA, UCAS-AOD, and HRSC2016—sufficiently corroborate the superiority of symmetry-aware enhancement schemes, which boast straightforward implementation and efficient inference deployment.

Abstract Augmented reality (AR) games, particularly those designed for head-mounted displays, have grown increasingly prevalent. However, most existing systems depend on pre-scanned, static environments and rely heavily on continuous tracking or marker-based solutions, which limit adaptability in dynamic physical spaces. This is particularly problematic for AR headsets and glasses, which typically follow the user’s head movement and cannot maintain a fixed, stationary view of the scene. Moreover, continuous scene observation is neither power-efficient nor practical for wearable devices, given their limited battery and processing capabilities. A persistent challenge arises when multiple identical objects are present in the environment–standard object tracking pipelines often fail to maintain consistent identities without uninterrupted observation or external sensors. These limitations hinder fluid physical-virtual interactions, especially in dynamic or occluded scenes where continuous tracking is infeasible. To address this, we introduce a novel optimization-based framework for re-identifying identical objects in AR scenes using only one partial egocentric observation frame captured by a headset. We formulate the problem as a label assignment task solved via integer programming, augmented with a Voronoi diagram-based pruning strategy to improve computational efficiency. This method reduces computation time by 50% while preserving 91% accuracy in simulated experiments. Moreover, we evaluated our approach in quantitative synthetic and quantitative real-world experiments. We also conducted three qualitative real-world experiments to demonstrate the practical utility and generalizability for enabling dynamic, markerless object interaction in AR environments. Our video demo is available at https://youtu.be/RwptEfLtW1U .

Unsupervised and semi-supervised clustering via density and distance-based label propagation and assignment

Background/Objectives: An impediment to successful drug discovery is the potential for off-target liabilities to eliminate otherwise promising candidates. As the drug discovery process is time-consuming and expensive, the use of artificial intelligence (AI) methods such as machine learning (ML) has drastically increased. It is invaluable to generate models that can quickly differentiate between successful and unsuccessful small-molecule drug candidates. Previous efforts established that molecular similarity could be used with other metrics to inform predictions of potential activity against a protein target. Similar methods were pursued here to combine similarity and machine learning for a collection of models called OPLE. Methods: Models were trained with proprietary and publicly available data to predict the likelihood of a given compound to be active against targets present in existing experimental SafetyScreen panels 18 and 44. Two-dimensional (2D) Tanimoto similarity from extended-connectivity fingerprints (ECFPs) and trained ML models were combined to obtain predictions. Results: Using all training data, a relationship between similarity and activity was established by fitting a probability assignment curve. Calibrated ML label assignment likelihoods were joined with the predictions from ECFP Tanimoto similarity to known active compounds using the belief theory formula, which maintains that activity prediction increases when both pieces of evidence support it. When assessing the performance of OPLE models for SafetyScreen 18 and 44 targets with external data from ChEMBL, more than 80% of the models had recall values greater than 0.8. This indicated favorable predictive ability to identify active molecules while limiting false negative predictions. Conclusions: Predicting and experimentally verifying safety liabilities is insightful at every stage of small-molecule drug discovery. This early detection tool can help project teams save resources that could be better deployed on series with no predicted or measured off-target liabilities.

The rapid development of Convolutional Neural Networks (CNNs) has markedly boosted the performance of object detection in remote sensing. Nevertheless, tiny objects typically account for an extremely small fraction of the total area in remote sensing images, rendering existing IoU-based or area-based evaluation metrics highly sensitive to minor pixel deviations. Meanwhile, classic detection models face inherent bottlenecks in efficiently mining discriminative features for tiny objects, leaving the task of tiny object detection in remote sensing images as an ongoing challenge in this field. To alleviate these issues, this paper proposes a tiny object detection method based on Normalized Gaussian Label Assignment and Multi-scale Hybrid Attention. Firstly, 2D Gaussian modeling is performed on the feature receptive field and the actual bounding box, using Normalized Bhattacharyya Distance for precise similarity measurement. Furthermore, a candidate sample quality ranking mechanism is constructed to select high-quality positive samples. Finally, a Multi-scale Hybrid Attention module is designed to enhance the discriminative feature extraction of tiny objects. The proposed method achieves 25.7% and 27.9% AP on the AI-TOD-v2 and VisDrone2019 datasets, respectively, significantly improving the detection capability of tiny objects in complex remote sensing scenarios.

In the field of precision agriculture, accurately detecting rice panicles is crucial for monitoring rice growth and managing rice production. To address the challenges posed by complex field backgrounds, including variety differences, variations across growth stages, background interference, and occlusion due to dense distribution, this study develops an improved YOLO11-based rice panicle detection model, termed DRPU-YOLO11. The model incorporates a task-oriented CSP-PGMA module in the backbone to enhance multi-scale feature extraction and provide richer representations for downstream detection. In the neck network, DySample and CGDown are adopted to strengthen global contextual feature aggregation and suppress background interference for small targets. Furthermore, fine-grained P2 level information is integrated with higher-level features through a cross-scale fusion module (CSP-ONMK) to improve detection robustness in dense and occluded scenes. In addition, the PowerTAL strategy adapts quality-aware label assignment to emphasize high-quality predictions during training. The experimental results based on a self-constructed dataset demonstrate that DRPU-YOLO11 significantly outperforms baseline models in rice panicle detection under complex field environments, achieving an accuracy of 82.5%. Compared with the baseline model YOLO11 and RT-DETR, the mAP50 increases by 2.4% and 5.0%, respectively. These results indicate that the proposed task-driven design provides a practical and high-precision solution for rice panicle detection, with potential applications in rice growth monitoring and yield estimation.

Existing object detection methods struggle to generalize across increasingly data domains while simultaneously adapting to the emergence of novel categories. To tackle this challenge, adaptive open-set object detection (AOOD) has been introduced, which employs supervised training on base categories within the source domain while enabling unsupervised adaptation to both base and novel categories in the target domain. However, existing AOOD approaches are still hindered by several limitations, including insufficient cross-domain feature representation, inter-category ambiguity in novel classes, and inherent feature bias toward the source domain. To overcome these issues, this paper proposes a category-level collaboration knowledge mining strategy designed to comprehensively exploit both inter-class and intra-class feature relationships across domains. Specifically, a clustering-based memory bank (CMB) is initially constructed to aggregate class prototype features, class auxiliary features, and intra-class disparity features, thereby embedding rich category-level knowledge into a unified memory structure. The CMB is iteratively updated through unsupervised clustering, which facilitates the modeling of intra-category relationships and enhances its capacity for cross-domain knowledge representation. Subsequently, a base-to-novel selection metric (BNSM) is designed to identify features corresponding to novel categories within the source domain by regulating the relationships between the novel categories and each base category. The selected features are then leveraged to initialize the object detector for the classification of novel categories. Finally, an adaptive feature assignment (AFA) strategy is introduced to transfer the learned category-level knowledge to the target domain, enabling the assignment of category labels to features. The memory bank is updated asynchronously with these assigned features to mitigate source domain bias. Extensive experiments conducted on diverse domain datasets demonstrate that the proposed method consistently outperforms state-of-the-art AOOD approaches, achieving performance gains of 1.1 to 5.5 mAP. Code is available at https://github.com/Jandsome/CCKM.

Dual Assignment of labels for end-to-end fully convolutional object detection

Robust object detection under varying weather conditions (e.g., rain, fog, and snow) presents significant challenges for industrial vision systems due to inherent visual degradations in manufacturing sites and outdoor facilities. While knowledge distillation offers promising potential by feature imitating and logit mimicking, existing methods face two critical limitations: first, inadequate mechanisms for effectively transferring localization capabilities in the presence of severe image degradation, and second, suboptimal strategies for identifying optimal distillation regions. To address these issues, we present a symmetric localization distillation loss based on the Jensen–Shannon divergence. Its mathematical characteristics, e.g., boundedness, symmetry, and gradient smoothness, enable robust preservation of spatial relationships and stabilize training processes. In addition, we present an adaptive label assignment strategy to select distillation regions, thus reducing the sparsity of positive samples during our knowledge distillation process. This work is the first one to apply knowledge distillation to object detection in inclement weather conditions. Extensive experiments on three challenging datasets show that our method improves the student model's object detection accuracy while maintaining its inference speed.

In recent years, significant progress has been made in remote sensing object detection task. However, the detection accuracy is usually degraded by the dense distribution and arbitrary orientation of objects. Besides, the small objects are prone to being missed detected. To address the above issues, a two-stage anchor-based oriented object detector has been proposed. First, a circumcircle representation has been proposed and incorporated into both proposal generation and refinement networks. The size and orientation of objects are characterized more accurately, thus boosting the detection accuracy of densely packed objects. Furthermore, a similarity analysis via Hellinger distance measure has been devised based on the Gaussian distribution of ground truth box and preset anchor. More high quality positive samples are correctly selected, hence the missed detections are reduced. Qualitative and quantitative experimental results demonstrate the superiority of the proposed detector.

Exploiting implicit knowledge for streaming perception object detection

To facilitate early detection of breast cancer, there is a need to develop risk prediction schemes that can prescribe personalized screening mammography regimens for women. In this study, we propose a new deep learning architecture called TRINet that implements time-decay attention to focus on recent mammographic screenings, as current models do not account for the relevance of newer images. We integrate radiomic features with an Attention-based Multiple Instance Learning (AMIL) framework to weigh and combine multiple views for better risk estimation. In addition, we introduce a continual learning approach with a new label assignment strategy based on bilateral asymmetry to make the model more adaptable to asymmetrical cancer indicators. Finally, we add a time-embedded additive hazard layer to perform dynamic, multi-year risk forecasting based on individualized screening intervals. We used two public datasets, namely 8528 patients from the American EMBED dataset and 8723 patients from the Swedish CSAW dataset in our experiments. Evaluation results on the EMBED test set show that our approach performs comparably with state-of-the-art models, achieving AUC scores of 0.851, 0.811, 0.796, 0.793, and 0.789 across 1-, 2-, to 5-year intervals, respectively. Our results underscore the importance of integrating temporal attention, radiomic features, time embeddings, bilateral asymmetry, and continual learning strategies, providing a more adaptive and precise tool for breast cancer risk prediction.

CLAIM: A knowledge-intensive, confidence-based label assignment and integration method for crowdsourced construction image data

FSS: Focusing on Suboptimal Samples for Detector-Agnostic Label Assignment in Object Detection

Semi-supervised multimodal contrastive regularization network for remote sensing hyperspectral and LiDAR classification

Abstract Object detection serves as a challenging yet crucial task in computer vision. Despite significant advancements, modern detectors remain struggling with task alignment between localization and classification. In this paper, Global Collaborative Learning (GCL) is introduced to address these challenges from often-overlooked perspectives. First, the essence of GCL is reflected in the label assignment of the detector. Adjusting the loss function to transform samples with strong localization yet weak classification into high-quality samples in both tasks, provides more effective training signals, enabling the model to capture key consistent features. Second, the spirit of GCL is embodied in the head design. By enabling global feature interaction within the decoupled head, the approach ensures that final predictions are made more comprehensively and robustly, thereby preventing the two independent branches from converging into suboptimal solutions for their respective tasks. Extensive experiments on the challenging MS COCO and CrowdHuman datasets demonstrate that the proposed GCL method substantially enhances performance and generalization capabilities.

The advancement of Artificial Intelligence (AI) in creative domains has sparked discussions regarding how listeners perceive and engage with AI-generated music. This study investigated listeners' emotions, perceptions, and attitudes toward AI-generated versus human-composed pop music. The study hypothesized that listeners would rate music labelled as AI-generated lower in terms of liking, quality, positive emotions, sensorial, imaginal, and experiential responses, as well as need for re-experience and purchase intentions, compared to music labelled as human-composed. Participants listened to eight AI-generated pop songs, four labelled as AI-generated and four labelled as human-composed. They then rated each song on various dimensions. To ensure a balanced design, label assignment on composer identity was fully randomized across both participants and songs. Contrary to the hypotheses, the participants rated pop songs labelled as AI-generated more highly in positive emotions, including happiness, interest, awe, and energy, compared to those labelled as human-composed. No significant differences were found between purported composer identity in the remaining dimensions. These results suggest that while the perception of AI authorship does influence listeners, the effects are primarily affective rather than sensorial, imaginal, experiential, or behavioural. Notably, considering that listeners rated pop songs labelled as AI-generated more positively in emotions, the findings imply that AI-generated music may be more readily accepted than previously assumed.