A-SelecT: Automatic Timestep Selection for Diffusion Transformer Representation Learning

TL;DR

This paper introduces A-SelecT, an automatic method for selecting the most informative timestep in Diffusion Transformers, improving training efficiency and representation quality for discriminative tasks.

Contribution

A-SelecT dynamically identifies the key timestep in DiT, eliminating exhaustive search and enhancing downstream discriminative performance.

Findings

A-SelecT improves classification accuracy on benchmarks.

A-SelecT enhances segmentation performance.

DiT with A-SelecT outperforms prior diffusion models.

Abstract

Diffusion models have significantly reshaped the field of generative artificial intelligence and are now increasingly explored for their capacity in discriminative representation learning. Diffusion Transformer (DiT) has recently gained attention as a promising alternative to conventional U-Net-based diffusion models, demonstrating a promising avenue for downstream discriminative tasks via generative pre-training. However, its current training efficiency and representational capacity remain largely constrained due to the inadequate timestep searching and insufficient exploitation of DiT-specific feature representations. In light of this view, we introduce Automatically Selected Timestep (A-SelecT) that dynamically pinpoints DiT's most information-rich timestep from the selected transformer feature in a single run, eliminating the need for both computationally intensive exhaustive timestep searching and suboptimal discriminative feature selection. Extensive experiments on classification and segmentation benchmarks demonstrate that DiT, empowered by A-SelecT, surpasses all prior diffusion-based attempts efficiently and effectively.