DiffTAD: Temporal Action Detection with Proposal Denoising Diffusion

TL;DR

DiffTAD introduces a generative diffusion-based approach for temporal action detection, transforming proposal generation into a denoising process that improves accuracy and convergence over traditional discriminative methods.

Contribution

This paper presents a novel diffusion-based framework for TAD, utilizing a Transformer decoder and a new inference acceleration technique, outperforming existing methods.

Findings

Achieves top performance on ActivityNet and THUMOS datasets.

Introduces a proposal denoising diffusion process for TAD.

Demonstrates faster training convergence with the proposed design.

Abstract

We propose a new formulation of temporal action detection (TAD) with denoising diffusion, DiffTAD in short. Taking as input random temporal proposals, it can yield action proposals accurately given an untrimmed long video. This presents a generative modeling perspective, against previous discriminative learning manners. This capability is achieved by first diffusing the ground-truth proposals to random ones (i.e., the forward/noising process) and then learning to reverse the noising process (i.e., the backward/denoising process). Concretely, we establish the denoising process in the Transformer decoder (e.g., DETR) by introducing a temporal location query design with faster convergence in training. We further propose a cross-step selective conditioning algorithm for inference acceleration. Extensive evaluations on ActivityNet and THUMOS show that our DiffTAD achieves top performance compared to previous art alternatives. The code will be made available at https://github.com/sauradip/DiffusionTAD.