PCA of Embeddings

1D projections encode quasi-sinusoidal signals from repetitions in videos (from QUVA Repetition dataset).

X-axis denotes time, Y-axis denotes 1st PCA of the embeddings, Red dot denotes the embedding of current frame