Pyramid With Super Resolution for In-the-Wild Facial Expression Recognition

Abstract

Facial Expression Recognition (FER) is a challenging task that improves natural humancomputer interaction. This paper focuses on automatic FER on a single in-the-wild (ITW) image. ITW images suffer real problems of pose, direction, and input resolution. In this study, we propose a pyramid with super-resolution (PSR) network architecture to solve the ITW FER task. We also introduce a prior distribution label smoothing (PDLS) loss function that applies the additional prior knowledge of the confusion about each expression in the FER task. Experiments on the three most popular ITW FER datasets showed that our approach outperforms all the state-of-the-art methods.

Vision-Based Road Rage Detection Framework in Automotive Safety Applications

“…The accuracy obtained with the DL architecture proposed in this work is in line with the accuracy reported in other scientific publications for both the RAF-DB dataset and the AffectNet dataset. Regarding the RAF-DB dataset, the difference in accuracy with the work that currently reports the highest accuracy is about 2.2%, but in [ 59 ] the accuracy reported for “anger” and “disgust” is lower than the accuracy obtained through the algorithmic pipeline proposed in this paper. On the other hand, the results reported in Table 7 show that the AffectNet dataset is the one containing facial expression images acquired in largely uncontrolled contexts, but even in this case our pipeline achieves a relatively lower accuracy (about 1.8%) compared to the current state of the art.…”

“…The limit of this dataset, however, is related to the fact that it contains only grey level images. Another important difference to headline is related to the computational cost of our approach with respect to the approach proposed for example in [ 59 ], in which the greater accuracies are reported. In fact, in order to integrate our system on a microcontroller, our pipeline includes deep learning architectures that work even in the absence of GPUs, and this is not the case with the architecture proposed in ref.…”

“…In fact, in order to integrate our system on a microcontroller, our pipeline includes deep learning architectures that work even in the absence of GPUs, and this is not the case with the architecture proposed in ref. [ 59 ] since, as shown by the authors, the experiments were performed on pc equipped with Ubuntu 18.04, with 32G RAM and using a GeForce RTX 2080 Ti GPU with 11G GPU RAM.…”

“…Accuracy (%) DLP-CNN [57] 74.20 FSN [60] 72.46 DDA loss [61] 79.71 ALT [62] 76.50 Separate loss [63] 77.25 PSR [59] 80.78 Our Method 78.57…”

Research on the Enhancement of Laser Radar Range Image Recognition Using a Super-Resolution Algorithm

“…In addition, SR methods based on deep learning have also made significant progress in recent years [ 15 , 16 ]. However, the current research on the role of the SR algorithm in recognition is focused on specific fields such as 2D face recognition [ 17 , 18 ]. To the best of our knowledge, no relevant study covers the role of the SR algorithm in range image recognition.…”