PointOdyssey: A Large-Scale Synthetic Dataset for Long-Term Point Tracking

TL;DR

PointOdyssey introduces a large-scale synthetic dataset designed for long-term point tracking, emphasizing naturalistic motion and diversity to improve tracking algorithms, and demonstrates that training on this dataset enhances performance on real-world benchmarks.

Contribution

The paper presents a novel synthetic dataset with extensive annotations and diversity for long-term point tracking, along with modifications to existing methods to leverage this data effectively.

Findings

Training on PointOdyssey improves tracking performance on real benchmarks.

The dataset's diversity enhances generalization of tracking algorithms.

Modified PIPs method benefits from increased temporal receptive field.

Abstract

We introduce PointOdyssey, a large-scale synthetic dataset, and data generation framework, for the training and evaluation of long-term fine-grained tracking algorithms. Our goal is to advance the state-of-the-art by placing emphasis on long videos with naturalistic motion. Toward the goal of naturalism, we animate deformable characters using real-world motion capture data, we build 3D scenes to match the motion capture environments, and we render camera viewpoints using trajectories mined via structure-from-motion on real videos. We create combinatorial diversity by randomizing character appearance, motion profiles, materials, lighting, 3D assets, and atmospheric effects. Our dataset currently includes 104 videos, averaging 2,000 frames long, with orders of magnitude more correspondence annotations than prior work. We show that existing methods can be trained from scratch in our dataset and outperform the published variants. Finally, we introduce modifications to the PIPs point tracking method, greatly widening its temporal receptive field, which improves its performance on PointOdyssey as well as on two real-world benchmarks. Our data and code are publicly available at: https://pointodyssey.com