Abstract:Deep learning has achieved outstanding accuracy in medical image segmentation, particularly for objects like organs or tumors with smooth boundaries or large sizes. Whereas, it encounters significant difficulties with objects that have zigzag boundaries or are small in size, leading to a notable decrease in segmentation effectiveness. In this context, using a loss function that incorporates smoothness and volume information into a model's predictions offers a promising solution to these shortcomings. In this work, we introduce an Adaptive Focal Loss (A-FL) function designed to mitigate class imbalance by down-weighting the loss for easy examples that results in up-weighting the loss for hard examples and giving greater emphasis to challenging examples, such as small and irregularly shaped objects. The proposed A-FL involves dynamically adjusting a focusing parameter based on an object's surface smoothness, size information, and adjusting the class balancing parameter based on the ratio of targeted area to total area in an image. We evaluated the performance of the A-FL using ResNet50-encoded U-Net architecture on the Picai 2022 and BraTS 2018 datasets. On the Picai 2022 dataset, the A-FL achieved an Intersection over Union (IoU) of 0.696 and a Dice Similarity Coefficient (DSC) of 0.769, outperforming the regular Focal Loss (FL) by 5.5% and 5.4% respectively. It also surpassed the best baseline Dice-Focal by 2.0% and 1.2%. On the BraTS 2018 dataset, A-FL achieved an IoU of 0.883 and a DSC of 0.931. The comparative studies show that the proposed A-FL function surpasses conventional methods, including Dice Loss, Focal Loss, and their hybrid variants, in IoU, DSC, Sensitivity, and Specificity metrics. This work highlights A-FL's potential to improve deep learning models for segmenting clinically significant regions in medical images, leading to more precise and reliable diagnostic tools.
Abstract:Neural Rendering representations have significantly contributed to the field of 3D computer vision. Given their potential, considerable efforts have been invested to improve their performance. Nonetheless, the essential question of selecting training views is yet to be thoroughly investigated. This key aspect plays a vital role in achieving high-quality results and aligns with the well-known tenet of deep learning: "garbage in, garbage out". In this paper, we first illustrate the importance of view selection by demonstrating how a simple rotation of the test views within the most pervasive NeRF dataset can lead to consequential shifts in the performance rankings of state-of-the-art techniques. To address this challenge, we introduce a unified framework for view selection methods and devise a thorough benchmark to assess its impact. Significant improvements can be achieved without leveraging error or uncertainty estimation but focusing on uniform view coverage of the reconstructed object, resulting in a training-free approach. Using this technique, we show that high-quality renderings can be achieved faster by using fewer views. We conduct extensive experiments on both synthetic datasets and realistic data to demonstrate the effectiveness of our proposed method compared with random, conventional error-based, and uncertainty-guided view selection.
Abstract:Heatmaps have been instrumental in helping understand deep network decisions, and are a common approach for Explainable AI (XAI). While significant progress has been made in enhancing the informativeness and accessibility of heatmaps, heatmap analysis is typically very subjective and limited to domain experts. As such, developing automatic, scalable, and numerical analysis methods to make heatmap-based XAI more objective, end-user friendly, and cost-effective is vital. In addition, there is a need for comprehensive evaluation metrics to assess heatmap quality at a granular level.
Abstract:This paper addresses a critical preliminary step in radar signal processing: detecting the presence of a radar signal and robustly estimating its bandwidth. Existing methods which are largely statistical feature-based approaches face challenges in electronic warfare (EW) settings where prior information about signals is lacking. While alternate deep learning based methods focus on more challenging environments, they primarily formulate this as a binary classification problem. In this research, we propose a novel methodology that not only detects the presence of a signal, but also localises it in the time domain and estimates its operating frequency band at that point in time. To achieve robust estimation, we introduce a compound loss function that leverages complementary information from both time-domain and frequency-domain representations. By integrating these approaches, we aim to improve the efficiency and accuracy of radar signal detection and parameter estimation, reducing both unnecessary resource consumption and human effort in downstream tasks.
Abstract:Neural radiance fields (NeRFs) have exhibited potential in synthesizing high-fidelity views of 3D scenes but the standard training paradigm of NeRF presupposes an equal importance for each image in the training set. This assumption poses a significant challenge for rendering specific views presenting intricate geometries, thereby resulting in suboptimal performance. In this paper, we take a closer look at the implications of the current training paradigm and redesign this for more superior rendering quality by NeRFs. Dividing input views into multiple groups based on their visual similarities and training individual models on each of these groups enables each model to specialize on specific regions without sacrificing speed or efficiency. Subsequently, the knowledge of these specialized models is aggregated into a single entity via a teacher-student distillation paradigm, enabling spatial efficiency for online render-ing. Empirically, we evaluate our novel training framework on two publicly available datasets, namely NeRF synthetic and Tanks&Temples. Our evaluation demonstrates that our DaC training pipeline enhances the rendering quality of a state-of-the-art baseline model while exhibiting convergence to a superior minimum.
Abstract:The fusion of vision and language has brought about a transformative shift in computer vision through the emergence of Vision-Language Models (VLMs). However, the resource-intensive nature of existing VLMs poses a significant challenge. We need an accessible method for developing the next generation of VLMs. To address this issue, we propose Zoom-shot, a novel method for transferring the zero-shot capabilities of CLIP to any pre-trained vision encoder. We do this by exploiting the multimodal information (i.e. text and image) present in the CLIP latent space through the use of specifically designed multimodal loss functions. These loss functions are (1) cycle-consistency loss and (2) our novel prompt-guided knowledge distillation loss (PG-KD). PG-KD combines the concept of knowledge distillation with CLIP's zero-shot classification, to capture the interactions between text and image features. With our multimodal losses, we train a $\textbf{linear mapping}$ between the CLIP latent space and the latent space of a pre-trained vision encoder, for only a $\textbf{single epoch}$. Furthermore, Zoom-shot is entirely unsupervised and is trained using $\textbf{unpaired}$ data. We test the zero-shot capabilities of a range of vision encoders augmented as new VLMs, on coarse and fine-grained classification datasets, outperforming the previous state-of-the-art in this problem domain. In our ablations, we find Zoom-shot allows for a trade-off between data and compute during training; and our state-of-the-art results can be obtained by reducing training from 20% to 1% of the ImageNet training data with 20 epochs. All code and models are available on GitHub.
Abstract:Aerial-ground person re-identification (Re-ID) presents unique challenges in computer vision, stemming from the distinct differences in viewpoints, poses, and resolutions between high-altitude aerial and ground-based cameras. Existing research predominantly focuses on ground-to-ground matching, with aerial matching less explored due to a dearth of comprehensive datasets. To address this, we introduce AG-ReID.v2, a dataset specifically designed for person Re-ID in mixed aerial and ground scenarios. This dataset comprises 100,502 images of 1,615 unique individuals, each annotated with matching IDs and 15 soft attribute labels. Data were collected from diverse perspectives using a UAV, stationary CCTV, and smart glasses-integrated camera, providing a rich variety of intra-identity variations. Additionally, we have developed an explainable attention network tailored for this dataset. This network features a three-stream architecture that efficiently processes pairwise image distances, emphasizes key top-down features, and adapts to variations in appearance due to altitude differences. Comparative evaluations demonstrate the superiority of our approach over existing baselines. We plan to release the dataset and algorithm source code publicly, aiming to advance research in this specialized field of computer vision. For access, please visit https://github.com/huynguyen792/AG-ReID.v2.
Abstract:Recent progress in semantic scene understanding has primarily been enabled by the availability of semantically annotated bi-modal (camera and lidar) datasets in urban environments. However, such annotated datasets are also needed for natural, unstructured environments to enable semantic perception for applications, including conservation, search and rescue, environment monitoring, and agricultural automation. Therefore, we introduce WildScenes, a bi-modal benchmark dataset consisting of multiple large-scale traversals in natural environments, including semantic annotations in high-resolution 2D images and dense 3D lidar point clouds, and accurate 6-DoF pose information. The data is (1) trajectory-centric with accurate localization and globally aligned point clouds, (2) calibrated and synchronized to support bi-modal inference, and (3) containing different natural environments over 6 months to support research on domain adaptation. Our 3D semantic labels are obtained via an efficient automated process that transfers the human-annotated 2D labels from multiple views into 3D point clouds, thus circumventing the need for expensive and time-consuming human annotation in 3D. We introduce benchmarks on 2D and 3D semantic segmentation and evaluate a variety of recent deep-learning techniques to demonstrate the challenges in semantic segmentation in natural environments. We propose train-val-test splits for standard benchmarks as well as domain adaptation benchmarks and utilize an automated split generation technique to ensure the balance of class label distributions. The data, evaluation scripts and pretrained models will be released upon acceptance at https://csiro-robotics.github.io/WildScenes.
Abstract:Seizure events may manifest as transient disruptions in movement and behavior, and the analysis of these clinical signs, referred to as semiology, is subject to observer variations when specialists evaluate video-recorded events in the clinical setting. To enhance the accuracy and consistency of evaluations, computer-aided video analysis of seizures has emerged as a natural avenue. In the field of medical applications, deep learning and computer vision approaches have driven substantial advancements. Historically, these approaches have been used for disease detection, classification, and prediction using diagnostic data; however, there has been limited exploration of their application in evaluating video-based motion detection in the clinical epileptology setting. While vision-based technologies do not aim to replace clinical expertise, they can significantly contribute to medical decision-making and patient care by providing quantitative evidence and decision support. Behavior monitoring tools offer several advantages such as providing objective information, detecting challenging-to-observe events, reducing documentation efforts, and extending assessment capabilities to areas with limited expertise. In this paper, we detail the foundation technologies used in vision-based systems in the analysis of seizure videos, highlighting their success in semiology detection and analysis, focusing on work published in the last 7 years. We systematically present these methods and indicate how the adoption of deep learning for the analysis of video recordings of seizures could be approached. Additionally, we illustrate how existing technologies can be interconnected through an integrated system for video-based semiology analysis. Finally, we discuss challenges and research directions for future studies.
Abstract:Radio signal recognition is a crucial function in electronic warfare. Precise identification and localisation of radar pulse activities are required by electronic warfare systems to produce effective countermeasures. Despite the importance of these tasks, deep learning-based radar pulse activity recognition methods have remained largely underexplored. While deep learning for radar modulation recognition has been explored previously, classification tasks are generally limited to short and non-interleaved IQ signals, limiting their applicability to military applications. To address this gap, we introduce an end-to-end multi-stage learning approach to detect and localise pulse activities of interleaved radar signals across an extended time horizon. We propose a simple, yet highly effective multi-stage architecture for incrementally predicting fine-grained segmentation masks that localise radar pulse activities across multiple channels. We demonstrate the performance of our approach against several reference models on a novel radar dataset, while also providing a first-of-its-kind benchmark for radar pulse activity segmentation.