David Fan

ORCID: 0000-0002-9217-5451
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Human Pose and Action Recognition
  • Video Analysis and Summarization
  • Advanced Vision and Imaging
  • Pregnancy and preeclampsia studies
  • Robotic Path Planning Algorithms
  • Neurotransmitter Receptor Influence on Behavior
  • Birth, Development, and Health
  • Multimodal Machine Learning Applications
  • Computer Graphics and Visualization Techniques
  • Neuroscience and Neuropharmacology Research
  • Robotic Locomotion and Control
  • Memory and Neural Mechanisms
  • Robotics and Sensor-Based Localization
  • Optical measurement and interference techniques
  • Reproductive System and Pregnancy
  • Advanced Image Processing Techniques
  • Prosthetics and Rehabilitation Robotics
  • Appendicitis Diagnosis and Management
  • Control and Dynamics of Mobile Robots
  • Advanced Data Compression Techniques
  • Receptor Mechanisms and Signaling
  • Animal Behavior and Welfare Studies
  • Autonomous Vehicle Technology and Safety
  • Diverticular Disease and Complications
  • Studies on Chitinases and Chitosanases

Amazon (United States)
2021-2024

St. George's University
2024

MemorialCare Health System
2024

Jet Propulsion Laboratory
2023

Duke University
2010-2022

University of Dayton
2014-2021

University of California, Santa Barbara
2020-2021

Southeast University
2021

Princeton University
2017-2020

Rutgers, The State University of New Jersey
2014-2015

Interval schedules of reinforcement are known to generate habitual behavior, the performance which is less sensitive revaluation earned reward and alterations in action-outcome contingency. Here we report results from experiments using different types interval mice assess effect uncertainty, time availability, on habit formation. After limited training, lever pressing under fixed (FI, low uncertainty) or random (RI, higher was devaluation, but with more extended animals trained RI became...

10.3389/fnint.2010.00017 article EN cc-by Frontiers in Integrative Neuroscience 2010-01-01

Scenes play a crucial role in breaking the storyline of movies and TV episodes into semantically cohesive parts. However, given their complex temporal structure, finding scene boundaries can be challenging task requiring large amounts labeled training data. To address this challenge, we present self-supervised shot contrastive learning approach (ShotCoL) to learn representation that maximizes similarity between nearby shots compared randomly selected shots. We show how apply our learned for...

10.1109/cvpr46437.2021.00967 article EN 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2021-06-01

Single-view 3D is the task of recovering properties such as depth and surface normals from a single image. We hypothesize that major obstacle to single-image data. address this issue by presenting Open Annotations Single Image Surfaces (OASIS), dataset for in wild consisting annotations detailed geometry 140,000 images. train evaluate leading models on variety tasks. expect OASIS be useful resource vision research. Project site: https://pvl.cs.princeton.edu/OASIS.

10.1109/cvpr42600.2020.00076 article EN 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2020-06-01

10.1109/wacv61041.2025.00540 article EN 2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2025-02-26

A major output nucleus of the basal ganglia is substantia nigra pars reticulata, which sends GABAergic projections to brainstem and thalamic nuclei. The (GABA) neurons are reciprocally connected with nearby dopaminergic neurons, project mainly ganglia, a set subcortical nuclei critical for goal-directed behaviors. Here we examined impact motivational states on activity GABA in reticulata neighboring (DA) compacta. Both types show short-latency bursts cue predicting food reward. As mice...

10.1371/journal.pone.0071598 article EN cc-by PLoS ONE 2013-08-06

Risk is a ubiquitous feature of the environment for most organisms, who must often choose between small and certain reward larger but less reward. To study choice behavior under risk in genetically well characterized species, we trained mice (C57BL/6) on discrete trial, concurrent-choice task which they two levers. Pressing one lever (safe choice) always followed by other (risky reward, only some trials. The overall payoff same both When were not food deprived, indifferent to risk, choosing...

10.1371/journal.pone.0025342 article EN cc-by PLoS ONE 2011-09-22

Aging | doi:10.18632/aging.100806. Zachary A. Kopp, Jo-Lin Hsieh, Andrew Li, William Wang, Dhelni T. Bhatt, Angela Lee, Sae Yeon Kim, David Fan, Veevek Shah, Emaad Siddiqui, Radhika Ragam, Kristen Park, Dev Ardeshna, Kunwoo Rachel Wu, Hardik Parikh, Ayush Yuh-Ru Lin, Yongkyu Park

10.18632/aging.100806 article ID cc-by Aging 2015-09-21

Several recent works have directly extended the image masked autoencoder (MAE) with random masking into video domain, achieving promising results. However, unlike images, both spatial and temporal information are important for understanding. This suggests that strategy is inherited from MAE less effective MAE. motivates design of a novel algorithm can more efficiently make use saliency. Specifically, we propose motion-guided (MGM) which leverages motion vectors to guide position each mask...

10.1109/iccv51070.2023.00517 article EN 2021 IEEE/CVF International Conference on Computer Vision (ICCV) 2023-10-01

Autonomous exploration of unknown environments with aerial vehicles remains a challenge, especially in perceptually degraded conditions. Dust, fog, or lack visual LiDAR-based features results severe difficulties for state estimation algorithms, which failure can be catastrophic. In this work, we show that it is indeed possible to navigate such conditions without any exteroceptive sensing by exploiting collisions instead treating them as constraints. To end, present novel contact-based...

10.48550/arxiv.1909.00079 preprint EN other-oa arXiv (Cornell University) 2019-01-01

This study aimed to develop a blood test for the prediction of pre-eclampsia (PE) early in gestation. We hypothesised that longitudinal measurements circulating adipokines and sphingolipids maternal serum over course pregnancy could identify novel prognostic biomarkers are predictive impending event PE gestation.Retrospective discovery confirmation.Maternity units from two US hospitals.Six previously published studies placental tissue (78 95 non-PE) were compiled genomic discovery, sera 15...

10.1136/bmjopen-2021-050963 article EN cc-by-nc BMJ Open 2021-11-01

Unilateral dopamine (DA) depletion produces ipsiversive turning behaviour, and the injection of DA receptor agonists can produce contraversive turning, but underlying mechanisms remain unclear. We conducted in vivo recording pharmacological optogenetic manipulations to study role striatal output behaviour. used a video-based tracking programme while single unit activity both putative medium spiny projection neurons (MSNs) fast-spiking interneurons (FSIs) dorsal striatum bilaterally. Our...

10.1111/ejn.15764 article EN European Journal of Neuroscience 2022-07-08

We present a purely client-side web-application, UBiT2 (User-friendly BioInformatics Tools), that provides installation-free, offline alignment, analysis, and visualization of RNA-sequencing as well qPCR data. Analysis modules were designed with single cell transcriptomic analysis in mind. Using just browser, users can perform standard analyses such quality control, filtering, hierarchical clustering, principal component differential expression gene set enrichment testing, more, all...

10.1101/118992 preprint EN cc-by-nc bioRxiv (Cold Spring Harbor Laboratory) 2017-03-22

Although ground robotic autonomy has gained widespread usage in structured and controlled environments, unknown off-road terrain remains a difficult problem. Extreme, off-road, unstructured environments such as undeveloped wilderness, caves, rubble pose unique challenging problems for autonomous navigation. To tackle these we propose an approach assessing traversability planning safe, feasible, fast trajectory real-time. Our approach, which name STEP (Stochastic Traversability Evaluation...

10.48550/arxiv.2103.02828 preprint EN other-oa arXiv (Cornell University) 2021-01-01

A 36-year-old female presented to the gynecology office eight weeks after placement of a ParaGard intrauterine device (IUD). Upon gynecologic examination, strings IUD were not found. Magnetic resonance imaging was performed which reported embedded in sigmoid colon. Initial diagnostic laparoscopy done without bowel preparation and revealed an within colon mesocolon. Colonoscopy did reveal any breach colonic lumen. second planned with robotic-assisted technique preparation. Intraoperative...

10.7759/cureus.62238 article EN Cureus 2024-06-12

As the scale of data and models for video understanding rapidly expand, handling long-form input in transformer-based presents a practical challenge. Rather than resorting to sampling or token dropping, which may result information loss, merging shows promising results when used collaboration with transformers. However, application processing is not trivial. We begin premise that should rely solely on similarity tokens; saliency tokens also be considered. To address this, we explore various...

10.48550/arxiv.2410.23782 preprint EN arXiv (Cornell University) 2024-10-31

Audio Description (AD) plays a pivotal role as an application system aimed at guaranteeing accessibility in multimedia content, which provides additional narrations suitable intervals to describe visual elements, catering specifically the needs of visually impaired audiences. In this paper, we introduce $\mathrm{CA^3D}$, pioneering unified Context-Aware Automatic that AD event scripts with precise locations long cinematic content. Specifically, $\mathrm{CA^3D}$ consists of: 1) Temporal...

10.48550/arxiv.2412.10002 preprint EN arXiv (Cornell University) 2024-12-13

Previous research has studied the task of segmenting cinematic videos into scenes and narrative acts. However, these studies have overlooked essential multimodal alignment fusion for effectively efficiently processing long-form (> 60min). In this paper, we introduce Multimodal alignmEnt aGgregation distillAtion (MEGA) long-video segmentation. MEGA tackles challenge by leveraging multiple media modalities. The method coarsely aligns inputs variable lengths different modalities with positional...

10.1109/iccv51070.2023.02132 article EN 2021 IEEE/CVF International Conference on Computer Vision (ICCV) 2023-10-01

Scenes play a crucial role in breaking the storyline of movies and TV episodes into semantically cohesive parts. However, given their complex temporal structure, finding scene boundaries can be challenging task requiring large amounts labeled training data. To address this challenge, we present self-supervised shot contrastive learning approach (ShotCoL) to learn representation that maximizes similarity between nearby shots compared randomly selected shots. We show how apply our learned for...

10.48550/arxiv.2104.13537 preprint EN cc-by-nc-nd arXiv (Cornell University) 2021-01-01

We present a methodology to explore the capabilities of an existing interface for controlling robotic arm with information extracted from brainwaves. Brainwaves are collected through use Emotiv EPOC headset. The headset utilizes electroencephalography (EEG) technology collect active brain signals. employ software suites classify thoughts subject representing specific actions. system then sends appropriate signal control arm. identified several actions mapping, implemented these chosen...

10.5220/0005157803390344 article EN cc-by-nc-nd 2014-01-01

Single-view 3D is the task of recovering properties such as depth and surface normals from a single image. We hypothesize that major obstacle to single-image data. address this issue by presenting Open Annotations Single Image Surfaces (OASIS), dataset for in wild consisting annotations detailed geometry 140,000 images. train evaluate leading models on variety tasks. expect OASIS be useful resource vision research. Project site: https://pvl.cs.princeton.edu/OASIS.

10.48550/arxiv.2007.13215 preprint EN other-oa arXiv (Cornell University) 2020-01-01

This paper explores a compression framework combining convolutional neural networks (CNNs) with traditional image codecs. Various activation functions, codecs, optimizers, and upscaling techniques are tested to obtain the best quality. While it was originally applied grayscale images, framework's performance 8-bit IR images is also explored.

10.1109/naecon49338.2021.9696409 article EN 2021-08-16
Coming Soon ...