FlashDepth: Real-time Streaming Video Depth Estimation at 2K Resolution

  • 2025-04-09 18:59:31
  • Gene Chou, Wenqi Xian, Guandao Yang, Mohamed Abdelfattah, Bharath Hariharan, Noah Snavely, Ning Yu, Paul Debevec
  • 0

Abstract

A versatile video depth estimation model should (1) be accurate andconsistent across frames, (2) produce high-resolution depth maps, and (3)support real-time streaming. We propose FlashDepth, a method that satisfies allthree requirements, performing depth estimation on a 2044x1148 streaming videoat 24 FPS. We show that, with careful modifications to pretrained single-imagedepth models, these capabilities are enabled with relatively little data andtraining. We evaluate our approach across multiple unseen datasets againststate-of-the-art depth models, and find that ours outperforms them in terms ofboundary sharpness and speed by a significant margin, while maintainingcompetitive accuracy. We hope our model will enable various applications thatrequire high-resolution depth, such as video editing, and onlinedecision-making, such as robotics.

 

Quick Read (beta)

loading the full paper ...