Projects

FIW Database

FIW is the largest and most comprehensive image DB for kinship recognition to date! Check out webpage for benchmarks, data, data challenges, and more.
Visit project page.

BFW and Fair Face Recognition

Bias in FR is a contemporary concern, justifiably. To have solutions we must establish ground-work such aa a baseline to characterize and rate systems per demographic. Visit GitHub for more info, source code, data downloads, and more.

Profiler: Face Synthesizer

Face frontalization system with one-to-many inputs (more faces = better results). For this, we proposed a Dual-Attention GAN For Large-Pose Face Frontalization! Published in 2020 AAAI!
Check out on Github!

PhD Dissertation: Defense PresentatioN

Title
Title: Automatic Face Understanding: Recognizing Families in Photos
Date / Time
Nov 24, 2020 at 2:00 PM
Committee Members
Prof. Yun Fu (Advisor), Prof. Octavia Camps, Prof. Sarah Ostadabbas

Laplace Landmark Localization (Fig. 1)

Abstract

Landmark localization in images and videos is a classic problem solved in various ways. Nowadays, with deep networks prevailing throughout machine learning, there are revamped interests in pushing facial landmark detectors to handle more challenging data. Most efforts use network objectives based on L1 or L2 norms, which have several disadvantages. First of all, the generated heat-maps trans- late to the locations of landmarks (i.e. confidence maps) from which predicted landmark locations (i.e. the means) get penalized without accounting for the spread: a high- scatter corresponds to low confidence and vice-versa. For this, we introduce a LaplaceKL objective that penalizes for low confidence. Another issue is a dependency on labeled data, which are expensive to obtain and susceptible to error. To address both issues, we propose an adversarial training framework that leverages unlabeled data to improve model performance. Our method claims state-of-the-art on all of the 300W benchmarks and ranks second-to-best on the An- notated Facial Landmarks in the Wild (AFLW) dataset. Furthermore, our model is robust with a reduced size: 1/8 the number of channels (i.e. 0.0398 MB) is comparable to the state-of-the-art in real-time on CPU. Thus, this work is of high practical value to real-life application.

Fig. 1: The proposed semi-supervised framework for landmarks localization. The labeled and unlabeled branched are marked with blue and red arrows, respectfully. Given an input image, G produces K heat-maps, one for each landmark. Labels are used to generate real heat-maps as ω(sl). G produces fake samples from the unlabeled data. Source images are concatenated on heat-maps and passed to D.

Other Projects

Families In the Wild (FIW) Image Dataset

1st large-scale image database for kinship recognition [project page]
Call For Papers! An ACM MM 2017 Data Challenge Workshop [challenge page]
NE CV Workshop 2016 @ BU [extended abstract, ppt]
ACM MM 2016 @ University of Amsterdam, [paper, poster]

Person Re-ID

Complex Event Detection in Videos