A5015610386- Advanced Neural Network Applications
- Medical Image Segmentation Techniques
- Domain Adaptation and Few-Shot Learning
- Retinal Imaging and Analysis
- Advanced Image and Video Retrieval Techniques
- Digital Imaging for Blood Diseases
- Handwritten Text Recognition Techniques
- Image and Object Detection Techniques
- COVID-19 diagnosis using AI
- Vehicle License Plate Recognition
- Image Processing and 3D Reconstruction
- AI in cancer detection
- Cutaneous Melanoma Detection and Management
- Data Management and Algorithms
- Advanced Memory and Neural Computing
- Semantic Web and Ontologies
- EEG and Brain-Computer Interfaces
- Species Distribution and Climate Change
- Radiomics and Machine Learning in Medical Imaging
- Neuroscience and Neural Engineering
- Genetic and phenotypic traits in livestock
- Wildlife Ecology and Conservation
- Optical Coherence Tomography Applications
RWTH Aachen University
2020-2025
In this paper, we present SasWOT, the first training-free Semantic segmentation Architecture Search (SAS) framework via an auto-discovery proxy. is widely used in many real-time applications. For fast inference and memory efficiency, Previous SAS seeks optimal segmenter by differentiable or RL Search. However, significant computational costs of these training-based limit their practical usage. To improve search explore route but empirically observe that existing zero-cost proxies designed on...
We analyze the capabilities of foundation models addressing tedious task generating annotations for animal tracking. Annotating a large amount data is vital and can be make-or-break factor robustness tracking model. Robustness particularly crucial in tracking, as accurate over long time horizons essential capturing behavior animals. However, additional using counterproductive, quality just important. Poorly annotated introduce noise inaccuracies, ultimately compromising performance accuracy...
Designing metrics for evaluating instance segmentation revolves around comprehensively considering object detection and accuracy. However, other important properties, such as sensitivity, continuity, equality, are overlooked in the current study. In this paper, we reveal that most existing have a limited resolution of quality. They only conditionally sensitive to change masks or false predictions. For certain metrics, score can drastically narrow range which could provide misleading...
We propose a neural network-based framework to optimize the perceptions simulated by in silico retinal implant model pulse2percept. The overall pipeline consists of trainable encoder, pre-trained and evaluator. encoder is U-Net, which takes original image outputs stimulus. also trained mimic biomimetic perceptual implemented evaluator shallow VGG classifier, with images. Based on 10,000 test images from MNIST dataset, we show that convolutional performs significantly better than trivial...
To date, most instance segmentation approaches are based on supervised learning that requires a considerable amount of annotated object contours as training ground truth. Here, we propose framework searches for the target shape prior. The prior model is learned with variational autoencoder only very limited data: In our experiments, few dozens patches from dataset, well purely synthetic shapes, were sufficient to achieve results en par methods full access data two out three cell datasets....
As a proposal-free approach, instance segmentation through pixel embedding learning and clustering is gaining more emphasis. Compared with bounding box refinement approaches, such as Mask R-CNN, it has potential advantages in handling complex shapes dense objects. In this work, we propose simple, yet highly effective, architecture for object-aware learning. A distance regression module incorporated into our to generate seeds fast clustering. At the same time, show that features learned by...
We present an object detection based approach to localize handwritten regions from documents, which initially aims enhance the anonymization during data transmission. The concatenated fusion of original and preprocessed images containing both printed texts notes or signatures are fed into convolutional neural network, where bounding boxes learned detect handwriting. Afterwards, can be processed (e.g. replaced with redacted signatures) conceal personally identifiable information (PII). This...