A5066363357- Domain Adaptation and Few-Shot Learning
- Image and Signal Denoising Methods
- Advanced Image Processing Techniques
- Image Processing Techniques and Applications
- Remote-Sensing Image Classification
- Radiomics and Machine Learning in Medical Imaging
- Remote Sensing and Land Use
- Respiratory viral infections research
- Natural Language Processing Techniques
- Multimodal Machine Learning Applications
- Speech and dialogue systems
- AI in cancer detection
- Sparse and Compressive Sensing Techniques
Indian Institute of Technology Bombay
2021-2024
We tackle the problem of few-shot image classification in context remote sensing hyperspectral images (HSIs). Due to difficulties collecting a large number labeled training samples, techniques hold much prominence general. One bottlenecks designing learning (FSL) systems arises from fact that model is likely overfit presence few samples and complex spectral feature distributions land-cover classes. To this end, we introduce stable prototypical network (SPN) for FSL by judiciously...
We tackle the few-shot open-set recognition (FSOSR) problem in context of remote sensing hyperspectral image (HSI) classification. Prior research on OSR mainly considers an empirical threshold class prediction scores to reject outlier samples. Further, recent endeavors HSI classification fail recognize outliers due 'closed-set' nature and fact that entire distributions are unknown during training. To this end, we propose optimize a novel calibration network (OCN) together with feature...
Single-source open-domain generalization (SS-ODG) addresses the challenge of labeled source domains with supervision during training and unlabeled novel target testing. The domain includes both known classes from samples previously unseen classes. Existing techniques for SS-ODG primarily focus on calibrating source-domain classifiers to identify open in domain. However, these methods struggle visually fine-grained open-closed data, often misclassifying as closed-set Moreover, relying solely...
Capturing high-resolution magnetic resonance (MR) images is a time consuming process, which makes it unsuitable for medical emergencies and pediatric patients. Low-resolution MR imaging, by contrast, faster than its counterpart, but compromises on fine details necessary more precise diagnosis. Super-resolution (SR), when applied to low-resolution images, can help increase their utility synthetically generating with little additional time. In this paper, we present an SR technique that based...
The study of vision-and-language navigation (VLN) has typically relied on expert trajectories, which may not always be available in real-world situations due to the significant effort required collect them. On other hand, existing approaches training VLN agents that go beyond data involve augmentations or online exploration can tedious and risky. In contrast, it is easy access large repositories suboptimal offline trajectories. Inspired by research reinforcement learning (ORL), we introduce...
Self-supervised learning (SSL) has emerged as a promising paradigm in medical imaging, addressing the chronic challenge of limited labeled data healthcare settings. While SSL shown impressive results, existing studies domain are often scope, focusing on specific datasets or modalities, evaluating only isolated aspects model performance. This fragmented evaluation approach poses significant challenge, models deployed critical settings must not achieve high accuracy but also demonstrate robust...
Single-source open-domain generalization (SS-ODG) addresses the challenge of labeled source domains with supervision during training and unlabeled novel target testing. The domain includes both known classes from samples previously unseen classes. Existing techniques for SS-ODG primarily focus on calibrating source-domain classifiers to identify open in domain. However, these methods struggle visually fine-grained open-closed data, often misclassifying as closed-set Moreover, relying solely...
Capturing high-resolution magnetic resonance (MR) images is a time consuming process, which makes it unsuitable for medical emergencies and pediatric patients. Low-resolution MR imaging, by contrast, faster than its counterpart, but compromises on fine details necessary more precise diagnosis. Super-resolution (SR), when applied to low-resolution images, can help increase their utility synthetically generating with little additional time. In this paper, we present SR technique that based...