Sung Wook Baik

In recent years, object detection has become a cornerstone of many computer vision applications, relying heavily on the availability of high-quality annotated datasets. However, even widely used benchmarks often suffer from annotation issues such as inaccurate bounding boxes, misclassified objects, and missing labels. These annotation errors, especially localization errors, can greatly affect the training and evaluation of detection models. In this survey, we provide a data-centric and comprehensive review of existing methods for identifying and analyzing errors in object detection datasets. We examine the main components of error detection workflows, including annotation error taxonomies and model-agnostic detection techniques. In addition, we develop a standardized categorization of annotation error types specific to object detection, providing a foundation for consistent analysis and comparison across studies. We also perform manual inspections of selected benchmark datasets to observe and quantify common annotation errors in practice. Moreover, the survey highlights the datasets used for evaluating error detection methods and compares their scope and inherent challenges. Finally, we summarize the types of annotation errors found in existing benchmarks and provide recommendations for future research to enhance dataset quality and reliability in object detection.

Brain tumors are among the most severe neurological disorders, where precise identification of tumor subregions from multimodal magnetic resonance imaging (MRI) is essential for diagnosis, treatment planning, and monitoring. Accurate delineation of tumor subregions from 3D MRI volumes plays a critical role in improving clinical decision-making and patient outcomes. Convolutional neural networks capture local anatomical details but struggle with global contextual reasoning, while transformer-based approaches enhance global modeling at the cost of high computational overhead. Recently, state-space models have emerged as efficient alternatives for long-sequence modeling in high-dimensional medical imaging. However, brain tumor segmentation remains challenging due to tumor heterogeneity, ambiguous boundaries, and long-range dependencies in 3D volumes, and state-space models remain underexplored for structured, boundary-sensitive segmentation in high-uncertainty regions. Therefore, we propose BURN-Mamba, a Boundary-aware Attention and Uncertainty-guided Refinement Network with Mamba, for robust 3D brain tumor segmentation. The framework employs Mamba-based encoders to capture long-range volumetric dependencies while preserving local spatial information. Boundary-aware skip connections enhance structural continuity, and a multimodal fusion mechanism integrates complementary information across MRI modalities. To further improve segmentation in uncertain regions, a diffusion-based refinement strategy progressively corrects prediction errors and sharpens boundary delineation. BURN-Mamba demonstrates competitive performance across multiple benchmark datasets. Across five independent experimental runs, the model achieves mean ± standard deviation Dice/HD95 values of 0.921 ± 0.006/3.90 ± 0.68 mm on BraTS 2021, 0.930 ± 0.007/3.77 ± 0.60 mm on BraTS 2023 Glioma, and 0.908 ± 0.009/25.06 ± 6.48 mm on BraTS 2023 Meningioma. Compared with state-of-the-art methods, the proposed framework achieves consistently strong segmentation performance across tumor subregions in terms of Dice similarity and HD95. These results demonstrate that combining state-space modeling with boundary-aware feature propagation and uncertainty-guided refinement provides an effective framework for precise 3D brain tumor segmentation.

In the realm of animal voice recognition, this work introduces AVRNet, a task-specific integration framework that combines separable convolutions, hierarchical deep supervision through auxiliary classifiers, skip connections, and a dual-pooling channel-spatial attention mechanism configured for spectrogram-based animal vocalization recognition. The combination is designed to provide a favorable trade-off between recognition accuracy and computational cost. Together, these modules enhance the interpretability, training efficiency, and overall performance of the model, making a significant contribution to the development of animal voice recognition technology. In the existing methods for animal voice recognition, researchers used attention mechanisms with average or max-pooling layers in the attention mechanisms. However, average pooling may overlook fine details by focusing on the global context, while max pooling may miss broader patterns by emphasizing only the most prominent features. Therefore, the proposed work incorporates both pooling strategies in the channel and spatial attention mechanisms. To further enhance performance while reducing computational cost, separable convolution is introduced in the spatial attention mechanism. The modified attention mechanisms are incorporated into the animal voice recognition network to capture the intricate nuances of animal sounds, which play a pivotal role in identifying distinct animal species based on vocalizations. To demonstrate the effectiveness of the animal voice recognition network, this work created a novel dataset encompassing a diverse set of different species. The performance of the animal voice recognition network was rigorously evaluated on two datasets: the newly created dataset (containing 4378 samples in 16 classes) and a dataset presented by EmreSasmaz (containing 875 samples in 10 classes). The quantitative and qualitative analysis, along with ablation studies, showed that the proposed model consistently outperformed state-of-the-art methods on both the EmreSasmaz and the newly created datasets. Additionally, the statistical analysis test further confirms the proposed model’s robustness and effectiveness. Thus, the proposed model is promising for applications in biodiversity monitoring, ecological research, and conservation efforts, and has the potential to contribute significantly to different domains in which there is a need for effective and efficient species recognition through vocalizations.

Connected Vision Systems (CVS) are transforming a variety of applications, including autonomous vehicles, smart cities, surveillance, and human-robot interaction. These systems harness multi-view multi-camera (MVMC) data to provide enhanced situational awareness through the integration of MVMC tracking, re-identification (Re-ID), and action understanding (AU). However, deploying CVS in real-world, dynamic environments presents a number of challenges, particularly in addressing occlusions, diverse viewpoints, and environmental variability. Existing surveys have focused primarily on isolated tasks such as tracking, Re-ID, and AU, often neglecting their integration into a cohesive system. These reviews typically emphasize single-view setups, overlooking the complexities and opportunities provided by multi-camera collaboration and multi-view data analysis. To the best of our knowledge, this survey is the first to offer a comprehensive and integrated review of MVMC that unifies MVMC tracking, Re-ID, and AU into a single framework. We propose a unique taxonomy to better understand the critical components of CVS, dividing it into four key parts: MVMC tracking, Re-ID, AU, and combined methods. We systematically arrange and summarize the state-of-the-art datasets, methodologies, results, and evaluation metrics, providing a structured view of the field's progression. Furthermore, we identify and discuss the open research questions and challenges, along with emerging technologies such as lifelong learning, privacy, and federated learning, that need to be addressed for future advancements. The paper concludes by outlining key research directions for enhancing the robustness, efficiency, and adaptability of CVS in complex, real-world applications. We hope this survey will inspire innovative solutions and guide future research toward the next generation of intelligent and adaptive CVS.