Research map / 3D vision / Depth and stereo
FlashDepth: Real-time Streaming Video Depth Estimation at 2K Resolution
Gene Chou, Wenqi Xian, Guandao Yang, Mohamed Abdelfattah, Bharath Hariharan, Noah Snavely, Ning Yu, Paul Debevec
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add UNet super-resolution on high-res input (ablation) | FlashDepth-L · Unreal4K | delta1: 0.566 -> 0.490 | Hurt |
| add UNet super-resolution on high-res input (ablation) | FlashDepth-L · Unreal4K | F1: 0.048 -> 0.034 | Hurt |
| add DAGF depth super-resolution (ablation) | FlashDepth-L · Unreal4K | delta1: 0.566 -> 0.377 | Hurt |
| add DAGF depth super-resolution (ablation) | FlashDepth-L · Unreal4K | FPS: 30 -> 7.7 | Hurt |
| use hybrid cross-attention model instead of bilinear upsampling (ablation) | FlashDepth-L · Unreal4K | F1: 0.048 -> 0.109 | Improved |
| use hybrid cross-attention model instead of bilinear upsampling (ablation) | FlashDepth-L · UrbanSyn | F1: 0.136 -> 0.185 | Improved |
| process frames natively at 2K instead of 518px | FlashDepth-L · Unreal4K | F1: 0.048 -> 0.143 | Improved |
| process frames natively at 2K instead of 518px | FlashDepth-L · Unreal4K | FPS: 30 -> 6.0 | Hurt |
| use hybrid with FlashDepth-L features instead of FlashDepth-S alone (ablation) | FlashDepth-S · UrbanSyn | delta1: 0.784 -> 0.862 | Improved |
| use large model instead of small model (ablation) | FlashDepth · ETH3D | delta1: 0.786 -> 0.875 | Improved |
| replace Mamba with Transformer recurrent module (ablation) | FlashDepth-L · Unreal4K | delta1: 0.566 -> 0.608 | Improved |
| replace Mamba with Transformer recurrent module (ablation) | FlashDepth-L · Sintel | delta1: 0.642 -> 0.615 | Hurt |
Closest papers by meaning
- Video Depth Anything: Consistent Depth Estimation for Super-Long Videos (CVPR 2025 · similarity 0.89)
- Buffer Anytime: Zero-Shot Video Depth and Normal from Image Priors (CVPR 2025 · similarity 0.88)
- Depth Any Video with Scalable Synthetic Data (ICLR 2025 · similarity 0.87)
- QuartDepth: Post-Training Quantization for Real-Time Depth Estimation on the Edge (CVPR 2025 · similarity 0.86)
- DepthCrafter: Generating Consistent Long Depth Sequences for Open-world Videos (CVPR 2025 · similarity 0.86)
- SharpDepth: Sharpening Metric Depth Predictions Using Diffusion Distillation (CVPR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
