Research map / Video / Video segmentation and tracking
DeDelayed: Deleting Remote Inference Delay via On-Device Correction
Dan Jacobellis, Mateen Ulhaq, Fabien Racapé, Hyomin Choi, Neeraja J. Yadwadkar
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 delay-aware future prediction training to the remote model (ablation) | remote video model · BDD100K val | mIoU: remote inference worse than local beyond 2 frames; predictive remote extends utility to 5 frames (167 ms) | Improved |
| fuse remote predictive features into the local model (DeDelayed) (ablation) | remote predictive model · BDD100K val, 167 ms remote delay | mIoU: 0.596 -> 0.641 | Improved |
| add remote predictive features to local model | local image model · BDD100K val, 167 ms remote delay | mIoU: 0.588 -> 0.641 (6.7%pt in text) | Improved |
| give remote model video context without temporal prediction (ablation) | remote image model · BDD100K val | mIoU | No clear effect |
| increase remote round-trip delay for conventional remote inference | remote image inference · BDD100K val | mIoU: 8%pt. mIoU lower than local at 167 ms | Hurt |
| increase local model delay from 0 ms to 33 ms | DeDelayed · BDD100K val, remote delay 0 ms | mIoU: 0.661 -> 0.624 | Hurt |
| stale or missing remote features | DeDelayed local model · BDD100K val | mIoU | No clear effect |
Closest papers by meaning
- Real-Time Video Inference on Edge Devices via Adaptive Model Streaming (ICCV 2021 · similarity 0.84)
- FlashDepth: Real-time Streaming Video Depth Estimation at 2K Resolution (ICCV 2025 · similarity 0.84)
- FASTER: Rethinking Real-Time Flow VLAs (NeurIPS 2026 · similarity 0.84)
- FlashDecoder: Real-Time Latent-to-Pixel Streaming Decoder with Transformers (CVPR 2026 · similarity 0.83)
- HiRED: Attention-Guided Token Dropping for Efficient Inference of High-Resolution Vision-Language Models (AAAI 2025 · similarity 0.83)
- LiBrA-Net: Lie-Algebraic Bilateral Affine Fields for Real-Time 4K Video Dehazing (NeurIPS 2026 · similarity 0.83)
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.
