A5024685173- Species Distribution and Climate Change
- Domain Adaptation and Few-Shot Learning
- Advanced Image and Video Retrieval Techniques
- Identification and Quantification in Food
- Wildlife Ecology and Conservation
- Animal Vocal Communication and Behavior
- Video Surveillance and Tracking Methods
- Marine animal studies overview
- Anomaly Detection Techniques and Applications
- Mobile Crowdsensing and Crowdsourcing
- Avian ecology and behavior
- Smart Agriculture and AI
- Multimodal Machine Learning Applications
- Genomics and Phylogenetic Studies
- Digital Imaging for Blood Diseases
- Music and Audio Processing
- Speech and Audio Processing
- Digital Media Forensic Detection
- Advanced Neural Network Applications
- Image Retrieval and Classification Techniques
- Environmental DNA in Biodiversity Studies
- Remote-Sensing Image Classification
- Advanced Clustering Algorithms Research
- IoT and Edge/Fog Computing
- Generative Adversarial Networks and Image Synthesis
University of Massachusetts Amherst
2023-2024
Cornell University
2018-2023
Amherst College
2023
Cornell Lab of Ornithology
2023
MacAulay-Brown (United States)
2023
California Institute of Technology
2015-2019
UC San Diego Health System
2014
University of California, San Diego
2014
Existing image classification datasets used in computer vision tend to have a uniform distribution of images across object categories. In contrast, the natural world is heavily imbalanced, as some species are more abundant and easier photograph than others. To encourage further progress challenging real conditions we present iNaturalist detection dataset, consisting 859,000 from over 5,000 different plants animals. It features visually similar species, captured wide variety situations, all...
We introduce tools and methodologies to collect high quality, large scale fine-grained computer vision datasets using citizen scientists - crowd annotators who are passionate knowledgeable about specific domains such as birds or airplanes. worked with domain experts NABirds, a new quality dataset containing 48,562 images of North American 555 categories, part annotations bounding boxes. find that significantly more accurate than Mechanical Turkers at zero cost. bird measure the popular like...
We propose an architecture for fine-grained visual categorization that approaches expert human performance in the classification of bird species. Our first computes estimate object's pose; this is used to compute local image features which are, turn, classification. The are computed by applying deep convolutional nets patches located and normalized pose. perform empirical study a number pose normalization schemes, including investigation higher order geometric warping functions. novel...
The world is long-tailed. What does this mean for computer vision and visual recognition? main two implications are (1) the number of categories we need to consider in applications can be very large, (2) training examples most small. Current recognition algorithms have achieved excellent classification accuracy. However, they require many reach peak performance, which suggests that long-tailed distributions will not dealt with well. We analyze question context eBird, a large fine-grained...
Recent progress in self-supervised learning has resulted models that are capable of extracting rich representations from image collections without requiring any explicit label supervision. However, to date the vast majority these approaches have restricted themselves training on standard benchmark datasets such as ImageNet. We argue fine-grained visual categorization problems, plant and animal species classification, provide an informative testbed for learning. In order facilitate this area...
Current human-in-the-loop fine-grained visual categorization systems depend on a predefined vocabulary of attributes and parts, usually determined by experts. In this work, we move away from that expert-driven attribute-centric paradigm present novel interactive classification system incorporates computer vision perceptual similarity metrics in unified framework. At test time, users are asked to judge relative between query image various sets images, these general queries do not require...
Accurate estimates of salmon escapement - the number fish migrating upstream to spawn are key data for conservation and fishery management. Existing methods counting using high-resolution imaging sonar hardware non-invasive compatible with computer vision processing. Prior work in this area has utilized object detection tracking based automated counting. However, these techniques remain inaccessible many deployment sites due limited compute connectivity field. We propose an alternative...
We introduce a method to greatly reduce the amount of redundant annotations required when crowdsourcing such as bounding boxes, parts, and class labels. For example, if two Mechanical Turkers happen click on same pixel location annotating part in given image-an event that is very unlikely occur by random chance-, it strong indication correct. A similar type confidence can be obtained single Turker happened agree with computer vision estimate. thus incrementally collect variable number worker...
Camera traps are a valuable tool for studying biodiversity, but research using this data is limited by the speed of human annotation. With vast amounts now available it imperative that we develop automatic solutions annotating camera trap in order to allow scale. A promising approach based on deep networks trained human-annotated images. We provide challenge dataset explore whether such generalize novel locations, since systems once and may be deployed operate automatically new locations...
This paper introduces the Tropel system which enables non-technical users to create arbitrary visual detectors without first annotating a training set. Our primary contribution is crowd active learning pipeline that seeded with only single positive example and an unlabeled set of images. We examine crowd's ability train given severely limited themselves. presents series experiments reveal relationship between worker training, consensus average precision trained by crowd-in-the-loop learning....
Abstract Animal migration is one of nature's most spectacular phenomena, but migratory animals and their journeys are imperilled across the globe. Migratory birds among well‐studied on Earth, yet relatively little known about in‐flight behaviour during nocturnal migration. Because many migrating bird species vocalize flight, passive acoustic monitoring shows great promise for facilitating widespread Here, we present Nighthawk, a deep learning model designed to detect identify vocalizations...
We introduce a method for efficiently crowdsourcing multiclass annotations in challenging, real world image datasets. Our is designed to minimize the number of human that are necessary achieve desired level confidence on class labels. It based combining models worker behavior with computer vision. general: it can handle large classes, labels come from taxonomy rather than flat list, and model dependence when workers see history previous annotations. may be used as drop-in replacement...
Estimating the geographical range of a species from sparse observations is challenging and important geospatial prediction problem. Given set locations where has been observed, goal to build model predict whether present or absent at any location. This problem long history in ecology, but traditional methods struggle take advantage emerging large-scale crowdsourced datasets which can include tens millions records for hundreds thousands species. In this work, we use Spatial Implicit Neural...
Advances in machine vision technology are rapidly enabling new and innovative uses within the field of biodiversity. Computers now able to use images identify tens thousands species across a wide range taxonomic groups real time, notably demonstrated by iNaturalist.org, which suggests IDs users (https://www.inaturalist.org/pages/computer_vision_demo) as they create observation records. Soon it will be commonplace detect video feeds or camera mobile device search for species-related content...