A Spitting Image: Modular Superpixel Tokenization in Vision Transformers
This work addresses the limitation of semantic-agnostic tokenization in vision transformers, offering a modular framework for researchers and practitioners, though it is incremental as it builds on existing ViT architectures.
The paper tackled the problem of grid-based tokenization in Vision Transformers by proposing a modular superpixel tokenization strategy, which significantly improves attribution faithfulness and enables pixel-level granularity in dense prediction tasks while maintaining classification performance.
Vision Transformer (ViT) architectures traditionally employ a grid-based approach to tokenization independent of the semantic content of an image. We propose a modular superpixel tokenization strategy which decouples tokenization and feature extraction; a shift from contemporary approaches where these are treated as an undifferentiated whole. Using on-line content-aware tokenization and scale- and shape-invariant positional embeddings, we perform experiments and ablations that contrast our approach with patch-based tokenization and randomized partitions as baselines. We show that our method significantly improves the faithfulness of attributions, gives pixel-level granularity on zero-shot unsupervised dense prediction tasks, while maintaining predictive performance in classification tasks. Our approach provides a modular tokenization framework commensurable with standard architectures, extending the space of ViTs to a larger class of semantically-rich models.