BYEL : Bootstrap on Your Emotion Latent

According to the problem of dataset construction cost for training in deep learning and the development of generative models, more and more researches are being conducted to train with synthetic data and to inference using real data. We propose emotion aware Self-Supervised Learning using ABAW's Learning Synthetic Data (LSD) dataset. We pre-train our method to LSD dataset as a self-supervised learning and then use the same LSD dataset to do downstream training on the emotion classification task as a supervised learning. As a result, a higher result(0.63) than baseline(0.5) was obtained.