DELTAv2: Accelerating Dense 3D Tracking

TL;DR

This paper introduces DELTAv2, a novel algorithm that significantly accelerates dense long-term 3D point tracking in videos by addressing computational bottlenecks with a coarse-to-fine strategy and optimized correlation computation, achieving 5-100x speedup.

Contribution

DELTAv2 presents a new tracking algorithm that reduces computational costs and increases speed while maintaining accuracy through a coarse-to-fine approach and learnable interpolation.

Findings

Achieves 5-100x speedup over existing methods.

Maintains state-of-the-art tracking accuracy.

Introduces a learnable interpolation module for trajectory initialization.

Abstract

We propose a novel algorithm for accelerating dense long-term 3D point tracking in videos. Through analysis of existing state-of-the-art methods, we identify two major computational bottlenecks. First, transformer-based iterative tracking becomes expensive when handling a large number of trajectories. To address this, we introduce a coarse-to-fine strategy that begins tracking with a small subset of points and progressively expands the set of tracked trajectories. The newly added trajectories are initialized using a learnable interpolation module, which is trained end-to-end alongside the tracking network. Second, we propose an optimization that significantly reduces the cost of correlation feature computation, another key bottleneck in prior methods. Together, these improvements lead to a 5-100x speedup over existing approaches while maintaining state-of-the-art tracking accuracy.