A5044978076- Advanced Image and Video Retrieval Techniques
- Advanced Vision and Imaging
- Human Pose and Action Recognition
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Multimodal Machine Learning Applications
- Video Surveillance and Tracking Methods
- Robotics and Sensor-Based Localization
- Medical Image Segmentation Techniques
- Anomaly Detection Techniques and Applications
- Visual Attention and Saliency Detection
- Developmental Biology and Gene Regulation
- Image Retrieval and Classification Techniques
- Genomics and Chromatin Dynamics
- Gene expression and cancer classification
- Image Processing Techniques and Applications
- Cell Image Analysis Techniques
- 3D Surveying and Cultural Heritage
- Visual perception and processing mechanisms
- Advanced Image Processing Techniques
- 3D Printing in Biomedical Research
- Machine Learning and Data Classification
- Diabetic Foot Ulcer Assessment and Management
- Image and Signal Denoising Methods
- Face recognition and analysis
UC Irvine Health
2013-2024
University of California, Irvine
2014-2023
Amazon (United States)
2021-2022
Amazon (Germany)
2021
University of California, Los Angeles
2020
California Institute of Technology
2019
University of California System
2016-2019
University of California, Berkeley
2002-2010
Jet Propulsion Laboratory
1999-2003
Berkeley College
2002
This paper presents a database containing 'ground truth' segmentations produced by humans for images of wide variety natural scenes. We define an error measure which quantifies the consistency between differing granularities and find that different human same image are highly consistent. Use this dataset is demonstrated in two applications: (1) evaluating performance segmentation algorithms (2) measuring probability distributions associated with Gestalt grouping factors as well statistics...
This paper investigates two fundamental problems in computer vision: contour detection and image segmentation. We present state-of-the-art algorithms for both of these tasks. Our detector combines multiple local cues into a globalization framework based on spectral clustering. segmentation algorithm consists generic machinery transforming the output any hierarchical region tree. In this manner, we reduce problem to that detection. Extensive experimental evaluation demonstrates our methods...
The goal of this work is to accurately detect and localize boundaries in natural scenes using local image measurements. We formulate features that respond characteristic changes brightness, color, texture associated with boundaries. In order combine the information from these an optimal way, we train a classifier human labeled images as ground truth. output provides posterior probability boundary at each location orientation. present precision-recall curves showing resulting detector...
Spectral graph theoretic methods have recently shown great promise for the problem of image segmentation. However, due to computational demands these approaches, applications large problems such as spatiotemporal data and high resolution imagery been slow appear. The contribution this paper is a method that substantially reduces requirements grouping algorithms based on spectral partitioning making it feasible apply them very problems. Our approach technique numerical solution eigenfunction...
We analyze the computational problem of multi-object tracking in video sequences. formulate using a cost function that requires estimating number tracks, as well their birth and death states. show global solution can be obtained with greedy algorithm sequentially instantiates tracks shortest path computations on flow network. Greedy algorithms allow one to embed pre-processing steps, such nonmax suppression, within algorithm. Furthermore, we give near-optimal based dynamic programming which...
We propose a generic grouping algorithm that constructs hierarchy of regions from the output any contour detector. Our method consists two steps, an oriented watershed transform (OWT) to form initial contours, followed by construction ultra-metric map (UCM) defining hierarchical segmentation. provide extensive experimental evaluation demonstrate that, when coupled high-performance detector, OWT-UCM produces state-of-the-art image segmentations. These segmentations can optionally be further...
Contours and junctions are important cues for perceptual organization shape recognition. Detecting locally has proved problematic because the image intensity surface is confusing in neighborhood of a junction. Edge detectors also do not perform well near junctions. Current leading approaches to junction detection, such as Harris operator, based on 2D variation signal. However, drawback this strategy that it confuses textured regions with We believe right approach detection should take...
Pooling second-order local feature statistics to form a high-dimensional bilinear has been shown achieve state-of-the-art performance on variety of fine-grained classification tasks. To address the computational demands high dimensionality, we propose represent covariance features as matrix and apply low-rank classifier. The resulting classifier can be evaluated without explicitly computing map which allows for large reduction in compute time well decreasing effective number parameters...
Many state-of-the-art approaches for object recognition reduce the problem to a 0-1 classification task. Such reductions allow one leverage sophisticated classifiers learning. These models are typically trained independently each class using positive and negative examples cropped from images. At test-time, various post-processing heuristics such as non-maxima suppression (NMS) required reconcile multiple detections within between different classes image. Though crucial good performance on...
We introduce a method to generate vectorial representations of visual classification tasks which can be used reason about the nature those and their relations. Given dataset with ground-truth labels loss function, we process images through "probe network" compute an embedding based on estimates Fisher information matrix associated probe network parameters. This provides fixed-dimensional task that is independent details such as number classes requires no understanding class label semantics....
We describe a latent approach that learns to detect actions in long sequences given training videos with only whole-video class labels. Our makes use of two innovations attention-modeling weakly-supervised learning. First, and most notably, our framework uses an attention model extract both foreground background frames whose appearance is explicitly modeled. Most prior works ignore the background, but we show modeling it allows system learn richer notion their temporal extents. Second,...
We introduce a differentiable, end-to-end trainable framework for solving pixel-level grouping problems such as instance segmentation consisting of two novel components. First, we regress pixels into hyper-spherical embedding space so that from the same group have high cosine similarity while those different groups below specified margin. analyze choice dimension and margin, relating them to theoretical results on problem distributing points uniformly sphere. Second, instances, utilize...
Abstract Heart rate is under the precise control of autonomic nervous system. However, wiring peripheral neural circuits that regulate heart poorly understood. Here, we develop a clearing-imaging-analysis pipeline to visualize innervation intact hearts in 3D and employed multi-technique approach map parasympathetic sympathetic mice. We identify cholinergic neurons noradrenergic an intrinsic cardiac ganglion stellate ganglia, respectively, project sinoatrial node. also report response...
The presence of occluders significantly impacts performance systems for object recognition. However, occlusion is typically treated as an unstructured source noise and explicit models have lagged behind those appearance shape. In this paper we describe a hierarchical deformable part model face detection keypoint localization that explicitly occlusions parts. proposed structure makes it possible to augment positive training data with large numbers synthetically occluded instances. This allows...
Figure–ground organization refers to the visual perception that a contour separating two regions belongs one of regions. Recent studies have found neural correlates figure–ground assignment in V2 as early 10–25 ms after response onset, providing strong support for role local bottom–up processing. How much information about is available from locally computed cues? Using large collection natural images, which neighboring were assigned relation by human observers, we quantified extent figural...
We formulate a layered model for object detection and image segmentation. describe generative probabilistic that composites the output of bank detectors in order to define shape masks explain appearance, depth ordering, labels all pixels an image. Notably, our system estimates both class instance labels. Building on previous benchmark criteria segmentation, we novel score evaluates evaluate PASCAL 2009 2010 segmentation challenge data sets show good test results with state-of-the-art...
We present a simple, efficient model for learning boundary detection based on random forest classifier. Our approach combines (1) clustering of training examples simple partitioning the space local edge orientations and (2) scale-dependent calibration individual tree output probabilities prior to multiscale combination. The resulting outperforms published results challenging BSDS500 benchmark. Further, large datasets our requires substantially less memory speeds up time by factor 10 over...