A5102831353- Visual Attention and Saliency Detection
- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- AI in cancer detection
- Video Surveillance and Tracking Methods
- Radiomics and Machine Learning in Medical Imaging
- Cell Image Analysis Techniques
- Mobile Crowdsensing and Crowdsourcing
- Face recognition and analysis
- Multimodal Machine Learning Applications
- Cancer Genomics and Diagnostics
- Face and Expression Recognition
- Medical Image Segmentation Techniques
- Human Pose and Action Recognition
- Image and Video Quality Assessment
- Video Analysis and Summarization
- Data Visualization and Analytics
- Visual perception and processing mechanisms
- Tactile and Sensory Interactions
- Anomaly Detection Techniques and Applications
- Gene expression and cancer classification
- Domain Adaptation and Few-Shot Learning
- Generative Adversarial Networks and Image Synthesis
- Bladder and Urothelial Cancer Treatments
- Obsessive-Compulsive Spectrum Disorders
PathAI (United States)
2020-2024
Critical Path Institute
2022-2024
The University of Texas at Austin
2011-2019
Xiamen University
2016
Manipal Academy of Higher Education
2008
We propose an end-to-end learning framework for segmenting generic objects in videos. Our method learns to combine appearance and motion information produce pixel level segmentation masks all prominent formulate this task as a structured prediction problem design two-stream fully convolutional neural network which fuses together unified framework. Since large-scale video datasets with segmentations are problematic, we show how bootstrap weakly annotated videos existing image recognition...
Conditional Random Fields (CRFs) can be used as a discriminative approach for simultaneous sequence segmentation and frame labeling. Latent-Dynamic (LDCRFs) incorporates hidden state variables within CRFs which model sub-structure motion patterns dynamics between labels. Motivated by the success of LDCRFs in gesture recognition, we propose framework automatic facial expression recognition from continuous video modeling temporal variations shapes using LDCRFs. We show that proposed...
We propose a semi-automatic method to obtain foreground object masks for large set of related images. develop stagewise active approach propagation: in each stage, we actively determine the images that appear most valuable human annotation, then revise estimates all unlabeled accordingly. In order identify that, once annotated, will propagate well other examples, introduce an selection procedure operates on joint segmentation graph over It prioritizes intervention those are uncertain and...
The mode of manual annotation used in an interactive segmentation algorithm affects both its accuracy and ease-of-use. For example, bounding boxes are fast to supply, yet may be too coarse get good results on difficult images, freehand outlines slower supply more specific, they overkill for simple images. Whereas existing methods assume a fixed form input no matter the image, we propose predict tradeoff between effort. Our approach learns whether graph cuts will succeed if initialized with...
We present a novel form of interactive video object segmentation where few clicks by the user helps system produce full spatio-temporal interest. Whereas conventional pipelines take user's initialization as starting point, we show value in taking lead even initialization. In particular, for given frame, precomputes ranked list thousands possible hypotheses (also referred to region proposals) using image and motion cues. Then, looks at top proposals, on boundary carve away erroneous ones....
Our interviews with people who have visual impairments show clothes shopping is an important activity in their lives. Unfortunately, web sites remain largely inaccessible. We propose design recommendations to address online accessibility issues reported by visually impaired study participants and implementation, which we call BrowseWithMe, these issues. BrowseWithMe employs artificial intelligence automatically convert a product page into structured representation that enables user...
Abstract While alterations in nucleus size, shape, and color are ubiquitous cancer, comprehensive quantification of nuclear morphology across a whole-slide histologic image remains challenge. Here, we describe the development pan-tissue, deep learning-based digital pathology pipeline for exhaustive detection, segmentation, classification utility this morphologic biomarker discovery. Manually-collected annotations were used to train an object detection segmentation model identifying nuclei,...
We propose an end-to-end learning framework for segmenting generic objects in both images and videos. Given a novel image or video, our approach produces pixel-level mask all "object-like" regions-even object categories never seen during training. formulate the task as structured prediction problem of assigning object/background label to each pixel, implemented using deep fully convolutional network. When applied model further incorporates motion stream, network learns combine appearance...
Autism Spectrum Disorders (ASDs), a neuerodevelopmental disability in children is cause of major concern. The with ASDs find it difficult to express and recognize emotions which makes hard for them interact socially. Conventional methods use medicinal means, special education behavioral analysis. They are not always successful usually expensive. There significant need develop technology based effective intervention cure. We propose an interactive game design uses modern computer vision...
Foreground object segmentation is a critical step for many image analysis tasks. While automated methods can produce high-quality results, their failures disappoint users in need of practical solutions. We propose resource allocation framework predicting how best to allocate fixed budget human annotation effort order collect higher quality segmentations given batch images and methods. The based on proposed prediction module that estimates the algorithm-drawn segmentations. demonstrate value...
We propose an end-to-end learning framework for segmenting generic objects in videos. Our method learns to combine appearance and motion information produce pixel level segmentation masks all prominent formulate this task as a structured prediction problem design two-stream fully convolutional neural network which fuses together unified framework. Since large-scale video datasets with segmentations are problematic, we show how bootstrap weakly annotated videos existing image recognition...
To make decisions about the long-term preservation and access of large digital collections, archivists gather information such as collections' contents, their organizational structure, file format composition. date, process analyzing a collection - from data gathering to exploratory analysis final conclusions has largely been conducted using pen paper methods. help analyze large-scale collections for archival purposes, we developed an interactive visual analytics application. The application...
ABSTRACT While alterations in nucleus size, shape, and color are ubiquitous cancer, comprehensive quantification of nuclear morphology across a whole-slide histologic image remains challenge. Here, we describe the development pan-tissue, deep learning-based digital pathology pipeline for exhaustive detection, segmentation, classification utility this morphologic biomarker discovery. Manually-collected annotations were used to train an object detection segmentation model identifying nuclei,...
As collections become larger in size, more complex structure and increasingly diverse composition, new approaches are needed to help curators assess digital files make decisions about their long-term preservation. We present research on the use of interactive visualization analyze file characterization information for purpose assessing preservation condition a vast collection electronic records. The case study contains over 1,000,000 formats arranged varied record structures groups....
We propose an end-to-end learning framework for segmenting generic objects in both images and videos. Given a novel image or video, our approach produces pixel-level mask all "object-like" regions---even object categories never seen during training. formulate the task as structured prediction problem of assigning object/background label to each pixel, implemented using deep fully convolutional network. When applied model further incorporates motion stream, network learns combine appearance...
We propose a hybrid framework for consistently producing high-quality object tracks by combining an automated tracker with little human input. The key idea is to tailor module each dataset intelligently decide when failing and so humans should be brought in re-localize continued tracking. Our approach leverages self-supervised learning on unlabeled videos learn tailored representation target that then used actively monitor its tracked region the fails. Since labeled data not needed, our can...