A5041862255- Generative Adversarial Networks and Image Synthesis
- Advanced Vision and Imaging
- Advanced Image Processing Techniques
- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- COVID-19 diagnosis using AI
- Cell Image Analysis Techniques
- Radiomics and Machine Learning in Medical Imaging
- Computer Graphics and Visualization Techniques
- Ultrasound in Clinical Applications
- Image Processing Techniques and Applications
- Single-cell and spatial transcriptomics
- Image Retrieval and Classification Techniques
- Medical Image Segmentation Techniques
- Image and Signal Denoising Methods
- AI in cancer detection
- Face recognition and analysis
- Brain Tumor Detection and Classification
- Image and Object Detection Techniques
- Microfluidic and Bio-sensing Technologies
- Phonocardiography and Auscultation Techniques
- Advanced Graph Neural Networks
- Domain Adaptation and Few-Shot Learning
- Digital Media Forensic Detection
- Visual Attention and Saliency Detection
Weizmann Institute of Science
2010-2025
Methods for super-resolution can be broadly classified into two families of methods: (i) The classical multi-image (combining images obtained at subpixel misalignments), and (ii) Example-Based (learning correspondence between low high resolution image patches from a database). In this paper we propose unified framework combining these methods. We further show how combined approach applied to obtain super as little single (with no database or prior examples). Our is based on the observation...
Large-scale text-to-image generative models have been a revolutionary breakthrough in the evolution of AI, synthesizing diverse images with highly complex visual concepts. However, pivotal challenge leveraging such for real-world content creation is providing users control over generated content. In this paper, we present new framework that takes text-to- image synthesis to realm image-to-image translation - given guidance and target text prompt as input, our method harnesses power...
Generative Adversarial Networks (GANs) typically learn a distribution of images in large image dataset, and are then able to generate new from this distribution. However, each natural has its own internal statistics, captured by unique patches. In paper we propose an "Internal GAN'' (InGAN) - image-specific GAN which trains on single input learns It is synthesize plethora significantly different sizes, shapes aspect-ratios all with the same patch-distribution (same "DNA'') as image....
We present a method for semantically transferring the visual appearance of one natural image to another. Specifically, our goal is generate an in which objects source structure are "painted" with their related target image. Our works by training generator given only single structure/appearance pair as input. To integrate semantic information into framework—a pivotal component tackling this task-our key idea leverage pre-trained and fixed Vision Transformer (ViT) model serves external prior....
Abstract Multiplexed imaging enables measurement of multiple proteins in situ, offering an unprecedented opportunity to chart various cell types and states tissues. However, classification, the task identifying type individual cells, remains challenging, labor-intensive, limiting throughput. Here, we present CellSighter, a deep-learning based pipeline accelerate classification multiplexed images. Given small training set expert-labeled images, CellSighter outputs label probabilities for all...
We study the use of deep features extracted from a pretrained Vision Transformer (ViT) as dense visual descriptors. observe and empirically demonstrate that such features, when extractedfrom self-supervised ViT model (DINO-ViT), exhibit several striking properties, including: (i) encode powerful, well-localized semantic information, at high spatial granularity, object parts; (ii) encoded information is shared across related, yet different categories, (iii) positional bias changes gradually...
Image manipulation dates back long before the deep learning era. The classical prevailing approaches were based on maximizing patch similarity between input and generated output. Recently, single-image GANs introduced as a superior more sophisticated solution to image tasks. Moreover, they offered opportunity not only manipulate given image, but also generate large diverse set of different outputs from single natural image. This gave rise new tasks, which are considered “GAN-only”. However,...
The generative AI revolution has recently expanded to videos. Nevertheless, current state-of-the-art video models are still lagging behind image in terms of visual quality and user control over the generated content. In this work, we present a framework that harnesses power text-to-image diffusion model for task text-driven editing. Specifically, given source target text-prompt, our method generates high-quality adheres text, while preserving spatial layout motion input video. Our is based...
This paper introduces a new formulation for discrete image labeling tasks, the Decision Tree Field (DTF), that combines and generalizes random forests conditional fields (CRF) which have been widely used in computer vision. In typical CRF model unary potentials are derived from sophisticated forest or boosting based classifiers, however, pairwise assumed to (1) simple parametric form with pre-specified fixed dependence on data, (2) be defined basis of small neighborhood. contrast, DTF, local...
Lung ultrasound (LUS) is a cheap, safe and non-invasive imaging modality that can be performed at patient bed-side. However, to date LUS not widely adopted due lack of trained personnel required for interpreting the acquired frames. In this work we propose framework training deep artificial neural networks LUS, which may promote broader use LUS. When using evaluate patient's condition, both anatomical phenomena (e.g., pleural line, presence consolidations), as well sonographic artifacts...
Image segmentation is a fundamental task in computer vision. Data annotation for training supervised methods can be labor-intensive, motivating unsupervised methods. Current approaches often rely on extracting deep features from pre-trained networks to construct graph, and classical clustering like k-means normalized-cuts are then applied as post-processing step. However, this approach reduces the high-dimensional information encoded pair-wise scalar affinities. To address limitation, study...
Clustering is a fundamental task in unsupervised learning. The focus of this paper the Correlation functional which combines positive and negative affinities between data points. contribution two fold: (i) Provide theoretic analysis functional. (ii) New optimization algorithms can cope with large scale problems (>100K variables) that are infeasible using existing methods. Our provides probabilistic generative interpretation for functional, justifies its intrinsic "model-selection"...
Given very few images containing a common object of interest under severe variations in appearance, we detect the and provide compact visual representation that object, depicted by binary sketch. Our algorithm is composed two stages: (i) Detect mutually (yet non-trivial) ensemble `self-similarity descriptors' shared all input images. (ii) Having found such ensemble, `invert' it to generate sketch which best represents this ensemble. This provides simple while eliminating background clutter...
Generative Adversarial Networks (GANs) typically learn a distribution of images in large image dataset, and are then able to generate new from this distribution. However, each natural has its own internal statistics, captured by unique patches. In paper we propose an "Internal GAN" (InGAN) - image-specific GAN which trains on single input learns It is synthesize plethora significantly different sizes, shapes aspect-ratios all with the same patch-distribution (same "DNA") as image....
We present DINO-Tracker -- a new framework for long-term dense tracking in video. The pillar of our approach is combining test-time training on single video, with the powerful localized semantic features learned by pre-trained DINO-ViT model. Specifically, simultaneously adopts DINO's to fit motion observations test while tracker that directly leverages refined features. entire trained end-to-end using combination self-supervised losses, and regularization allows us retain benefit from...
Abstract Multiplexed imaging enables measurement of multiple proteins in situ , offering an unprecedented opportunity to chart various cell types and states tissues. However, classification, the task identifying type individual cells, remains challenging, labor-intensive, limiting throughput. Here, we present CellSighter, a deep-learning based pipeline accelerate classification multiplexed images. Given small training set expert-labeled images, CellSighter outputs label probabilities for all...
Abstract Understanding tissue structure and function requires tools that quantify the expression of multiple proteins at single-cell resolution while preserving spatial information. Current imaging technologies use a separate channel for each individual protein, inherently limiting their throughput scalability. Here, we present CombPlex (COMBinatorial multiPLEXing), combinatorial staining platform coupled with an algorithmic framework to exponentially increase number can be measured from C...