Propagate Yourself: Exploring Pixel-Level Consistency for Unsupervised Visual Representation Learning

Understand

Contrastive learning methods for unsupervised visual representation learning have reached remarkable levels of transfer performance.

• We argue that the power of contrastive learning has yet to be fully unleashed, as current methods are trained only on instance-level pretext tasks, leading to representations that may be sub-optimal for downstream tasks requiring dense pixel predictions.
• In this paper, we introduce pixel-level pretext tasks for learning dense feature representations.
• The first task directly applies contrastive learning at the pixel level.