Order Regularization on Ordinal Loss for Head Pose, Age and Gaze Estimation

نویسندگان

چکیده

Ordinal loss is widely used in solving regression problems with deep learning technologies. Its basic idea to convert classification while preserving the natural order. However, order constraint enforced only by ordinal label implicitly, leading real output values not strictly It causes network learn separable feature rather than discriminative feature, and possibly overfit on training set. In this paper, we propose regularization loss, which makes outputs explicitly constraining classifiers The proposed method contains two parts, i.e. similar-weights constraint, reduces ineffective space between classifiers, differential-bias enforces decision planes enhances discrimination power of classifiers. Experimental results show that our boosts performance original various such as head pose, age, gaze estimation, significant error reduction around 5%. Furthermore, outperforms state art all these tasks, gain 14.4%, 2.2% 6.5% age estimation respectively.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Gaze-Tracking Based on Head-Pose Estimation using Depth Camera

A gaze is an indicator what people focused on. The research purpose is to implement a gaze tracking application so that people know which part of a wide view target is the most interesting. This research uses Kinect Xbox 360 as a tracking device and Visual Studio 2013 as development environment (IDE). The research can be applied in business purpose, such as in advertisement area. This research ...

متن کامل

A Head Pose-free Approach for Appearance-based Gaze Estimation

(a) (b) (c) Figure 1: 2-D illustration of relationship between gaze direction and head pose. (a) Under a fixed head pose (r0, t0), gaze direction α can be estimated from appearance. (b) To obtain α under another head pose (r̂, t̂), the estimated α ′ should be corrected because of captured eye appearance distortion. (c) Under head pose (r̂, t̂), gaze direction under WCS should be further compensated...

متن کامل

Higher Order Support Tensor Regression for Head Pose Estimation

In this paper, we exploit the advantages of tensor representations and propose a Supervised Multilinear Learning Model for regression. The model is based on the Canonical (CANDECOMP)/Parallel Factors (PARAFAC) decomposition of tensors of multiple modes and allows the simultaneous projection of an input tensor to more than one discriminative directions along each mode. These projection weights a...

متن کامل

A Comprehensive Head Pose and Gaze Database

Within the past decade, many computational approaches have been developed to estimate gaze directions of persons based on their facial appearance. Most researchers used common face datasets with only a limited representation of different head poses to train and verify their algorithms. Moreover, in most datasets, faces have neither a defined gaze direction, nor do they incorporate different com...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

ژورنال

عنوان ژورنال: Proceedings of the ... AAAI Conference on Artificial Intelligence

سال: 2021

ISSN: ['2159-5399', '2374-3468']

DOI: https://doi.org/10.1609/aaai.v35i2.16240