Sequence tokens of a class-token-only contrastively trained ViT-1D carry usable beat, chord, and onset information that emerges during training without any frame-level supervision.
Applying NT-Xent loss only to the class token in a lightweight ViT-1D surprisingly enables sequence tokens to handle local tasks while con- tributing to global ones
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Emergent musical properties of a transformer under contrastive self-supervised learning
Sequence tokens of a class-token-only contrastively trained ViT-1D carry usable beat, chord, and onset information that emerges during training without any frame-level supervision.