SSR-Net: A Compact Soft Stagewise Regression Network for Age Estimation

Abstract

This paper presents a novel CNN model called Soft Stagewise Regression Network (SSR-Net) for age estimation from a single image with a compact model size. Inspired by DEX, we address age estimation by performing multi-class classification and then turning classification results into regression by calculating the expected values. SSR-Net takes a coarse-to-fine strategy and performs multi-class classification with multiple stages. Each stage is only responsible for refining the decision of its previous stage for more accurate age estimation. Thus, each stage performs a task with few classes and requires few neurons, greatly reducing the model size. For addressing the quantization issue introduced by grouping ages into classes, SSR-Net assigns a dynamic range to each age class by allowing it to be shifted and scaled according to the input face image. Both the multi-stage strategy and the dynamic range are incorporated into the formulation of soft stagewise regression. A novel network architecture is proposed for carrying out soft stagewise regression. The resultant SSR-Net model is very compact and takes only 0.32 MB. Despite its compact size, SSR-Net’s performance approaches those of the state-of-the-art methods whose model sizes are often more than 1500× larger.

C3AE: Exploring the Limits of Compact Model for Age Estimation

“…The image with high resolution contains rich local information, in return one with low resolution may contain global and scene information. Other than selecting one aligned facial center in SSR [38], we crop face centers with three granularity levels, as shown in Fig. 2, then fed them into the shared CNN network.…”

“…We study age estimation on three datasets: IMDB-WIKI [29], Morph II [28] and FG-NET [5]. We follow the conventions in the literature SSR [38], DEX [29] and Hot [29], WIKI-IMDB are used for pre-training and the ablation study. Because Morph II is the most popular and large benchmark for age estimation, we choose it for ablation studies.…”

“…However, it is not suitable for the performance evaluation on the age estimation because it contains much more noise. Thus, following previous works, e.g., SSR [38] and DEX [29], we utilize IMDB-WIKI only for pre-training.…”

“…Following SSR [38] and DEX [29], the model is firstly pre-trained on the IMDB and WIKI dataset, and is with size of 64 × 64 × 3. In all the experiments, Adam optimizer is employed.…”

“…The learning rate is decreased by a factor of the regression value with patience epochs 20 on the change value of 0.0005. Following SSR [38], the evaluation criteria is mean absolute value (MAE). The factor α in Eq.…”