Fig. 2

Data representation for each method. For easier visualization, temporal resolution is set as two time-steps per quarter note. a Musical input data. b Vectorial representation used in Anticipation-RNN [11] and Music InpaintNet [2]. Data is represented as a temporal sequence array of t time steps. The token “_” represents the state of holding a note. c Representation used in Music SketchNet [10]. The input is separated (“factorized”) into pitch and rhythm dimensions. d Representation used in VLI [9]. It captures each note event and groups them into a list of measures