Dear all,
you are cordially invited to join the master's thesis defense of Deepika Arneja on 13.09.22 at 12:00. You can join virtually via Zoom as follows:
Link:https://uni-bamberg.zoom.us/j/99191342076
Meeting-ID: 991 9134 2076
Passwort: 7Y^b@?
Deepika Arneja will defend her thesis Landmark-based Classication of Facial Expressions with a Spatio-Temporal Attention Graph Neural Network.
Abstract:
Facial expressions are one of the most potent ways to present emotion and communication. Decoding these emotional expressions has been an active research area. Facial expression recognition analyses facial expressions from images and videos to decode an individual’s emotional state. This master’s thesis presents
a dynamic graph-based spatial-temporal attention model for facial expression recognition. The proposed model considers spatio-temporal motion patterns in video frames. The key idea is to extract facial landmarks from the dataset in use and construct a fully-connect graph from the facial landmark skeleton. Structures and dynamics of facial landmark skeletons are learned automatically via a self-attention mechanism that performs in both spatial and temporal domains. The Actor Study dataset is used for this thesis, and anger, happiness,
fear, disgust, and sadness are the five emotions used for training our model. Extensive experiments are conducted on the Actor Study dataset to evaluate the performance and to explain the predictions made by the proposed model.
a dynamic graph-based spatial-temporal attention model for facial expression recognition. The proposed model considers spatio-temporal motion patterns in video frames. The key idea is to extract facial landmarks from the dataset in use and construct a fully-connect graph from the facial landmark skeleton. Structures and dynamics of facial landmark skeletons are learned automatically via a self-attention mechanism that performs in both spatial and temporal domains. The Actor Study dataset is used for this thesis, and anger, happiness,
fear, disgust, and sadness are the five emotions used for training our model. Extensive experiments are conducted on the Actor Study dataset to evaluate the performance and to explain the predictions made by the proposed model.
The presentation and discussion will be held in English.