Dynamic Token Pruning in Plain Vision Transformers for Semantic Segmentation

TL;DR

This paper introduces a Dynamic Token Pruning method for vision transformers in semantic segmentation, reducing computational costs by 20-35% through early token exit based on difficulty, inspired by human coarse-to-fine segmentation.

Contribution

It proposes a novel dynamic token pruning approach that allows early exit of easy tokens in vision transformers for semantic segmentation, maintaining accuracy while reducing computation.

Findings

Reduces computational cost by 20-35% on average.

Maintains segmentation accuracy despite pruning.

Operates dynamically based on input difficulty.

Abstract

Vision transformers have achieved leading performance on various visual tasks yet still suffer from high computational complexity. The situation deteriorates in dense prediction tasks like semantic segmentation, as high-resolution inputs and outputs usually imply more tokens involved in computations. Directly removing the less attentive tokens has been discussed for the image classification task but can not be extended to semantic segmentation since a dense prediction is required for every patch. To this end, this work introduces a Dynamic Token Pruning (DToP) method based on the early exit of tokens for semantic segmentation. Motivated by the coarse-to-fine segmentation process by humans, we naturally split the widely adopted auxiliary-loss-based network architecture into several stages, where each auxiliary block grades every token's difficulty level. We can finalize the prediction of easy tokens in advance without completing the entire forward pass. Moreover, we keep $k$ highest confidence tokens for each semantic category to uphold the representative context information. Thus, computational complexity will change with the difficulty of the input, akin to the way humans do segmentation. Experiments suggest that the proposed DToP architecture reduces on average $20\% - 35\%$ of computational cost for current semantic segmentation methods based on plain vision transformers without accuracy degradation.