المفاهيم الأساسية
The author proposes a Deep Multi-Task Learning approach for estimating ordinal and nominal attributes of faces, optimizing weights through homoskedastic uncertainty.
الملخص
The paper introduces a framework for joint estimation of ordinal and nominal face attributes using multi-task learning. By sharing low-level parameters and designing separate classifiers, the approach simplifies the task of ordinal attribute estimation. The use of homoskedastic uncertainty to optimize loss weights among multiple tasks is a key contribution. Experimental results demonstrate superior performance compared to existing methods in terms of accuracy on benchmarks with multiple face attributes. The proposed approach is feasible for edge systems, addressing bias issues in face attribute estimation.
الإحصائيات
"Experimental results on benchmarks with multiple face attributes show that the proposed approach has superior performance compared to state of the art."
"The total number of training epochs was set to 80, the learning rate was set to 0.001, the weight decay value was set to 0.0005, batch size to 32."
"The accuracy of each attribute recognition on UTKFace is listed in Table II."
"For ordinal attribute estimation task, we computed the Mean Square Error (MSE) and Mean Absolute Error (MAE) for the age estimation task."
"The confusion matrix for face attributes along intersections of all attributes on the UTKFace benchmark is illustrated in Figure 6."