Yearbook Age Prediction (CS395T)
نویسنده
چکیده
We tackle the yearbook ’Face To Age’ challenge to predict the year that a given yearbook photo was taken in. We experiment with three variants of the VGG19 architecture [4], the AlexNet architecture [2], as well as different loss functions and an extensive effort to reduce overfitting. A variant of the AlexNet architecture, that has 2048 neurons in the fully connected layers and is ensembled over 5 models, gives us the best results, with a mean L1 norm distance of 5.54.
منابع مشابه
Image Prediction: Faces to Year Deep Learning Seminar (CS395T)
We finetune pretrained AlexNet and 16-layer VGGNet models over the yearbook dataset, predicting years in which the input images were taken. We discuss our approach and findings with the best performing model (L1 distance of 5.48 or 5% accuracy over the validation set).
متن کاملSIPRI Yearbook 2014 - Armaments, Disarmament and International Security.
sipri yearbook 2014 armaments disarmament and international security ebook, sipri yearbook 2014 armaments disarmament and international security pdf, sipri yearbook 2014 armaments disarmament and international security doc and sipri yearbook 2014 armaments disarmament and international security epub for sipri yearbook 2014 armaments disarmament and international security read online or sipri ye...
متن کاملEducation and consumer informatics.
OBJECTIVES To evaluate the extent to which the Internet is accessed for health information and perceived as useful to varying groups classified primarily according to age. METHOD Synopsis of the articles on education and consumer health informatics selected for the IMIA Yearbook of Medical Informatics 2010. RESULTS A growing number of individuals are actively seeking health information thro...
متن کاملOn the age dynamics of learned societies-taking the example of the Austrian Academy of Sciences.
In a hierarchical organisation of stable size the annual intake is strictly determined by the number of deaths and a statutory retirement age (if there is one). In this paper we reconstruct the population of the Austrian Academy of Sciences from 1847 to 2005. For the Austrian Academy of Sciences we observe a shift of its age distribution towards older ages, which on the one hand is due to risin...
متن کاملSensors, medical image and signal processing. Findings from the section on sensor, signal and imaging informatics.
OBJECTIVES To summarize current excellent research in the field of sensor, signal and imaging informatics. METHOD Synopsis of the articles selected for the IMIA Yearbook 2007. RESULTS The selection process for this yearbook section "Sensor, signal and imaging informatics" results in five excellent articles, representing research in four different nations. Papers from the fields of brain mac...
متن کامل