Handling ambiguous annotations for facial expression recognition in the wild

Gera, Darshan; Vikas, Gupta; Balasubramanian, Siva

doi:10.1145/3490035.3490289

Cited by 4 publications

(1 citation statement)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The difference between FER and FER+ is that the FER+ is the new relabelled dataset by ten crowd-sourced taggerregis, as can be seen in Fig. 2, which proved better quality ground truth for still image emotion than the original FER labels [14]. Ten taggers for each image enable researchers to estimate an emotion probability distribution per face, and this allows for constructing algorithms that produce statistical distributions or multi-label outputs instead of the conventional single-label output.…”

Section: A the Datasetmentioning

confidence: 99%

Facial Expression Recognition Using Convolutional Neural Network with Attention Module

Khoirullah

Yudistira²,

Bachtiar³

2022

JOIV : Int. J. Inform. Visualization

View full text Add to dashboard Cite

Human Activity Recognition (HAR) is an introduction to human activities that refer to the movements performed by an individual on specific body parts. One branch of HAR is human emotion. Facial emotion is vital in human communication to help convey emotional states and intentions. Facial Expression Recognition (FER) is crucial to understanding how humans communicate. Misinterpreting Facial Expressions can lead to misunderstanding and difficulty reaching a common ground. Deep Learning can help in recognizing these facial expressions. To improve the probation of Facial Expressions Recognition, we propose ResNet attached with an Attention module to push the performance forward. This approach performs better than the standalone ResNet because the localization and sampling grid allows the model to learn how to perform spatial transformations on the input image. Consequently, it improves the model's geometric invariance and picks up the features of the expressions from the human face, resulting in better classification results. This study proves the proposed method with attention is better than without, with a test accuracy of 0.7789 on the FER dataset and 0.8327 on the FER+ dataset. It concludes that the Attention module is essential in recognizing Facial Expressions using a Convolutional Neural Network (CNN). Advice for further research first, add more datasets besides FER and FER+, and second, add a Scheduler to decrease the learning rate during the training data.

show abstract

Section: A the Datasetmentioning

confidence: 99%