AI Paper Atlas

Research map / Video / Video segmentation and tracking

Real-Time Video Inference on Edge Devices via Adaptive Model Streaming

Mehrdad Khani, Pouya Hamadanian, Arash Nasr-Esfahany, Mohammad Alizadeh

ICCV 2021 · Video › Video segmentation and tracking · approach: Video semantic and panoptic segmentation · 60 citations (OpenAlex)

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 changedWhereMeasuredEffect
add continual online adaptation instead of no customizationDeepLabV3-MobileNetV2 · Four video datasetsmIoU: 0.4--17.8% (8.3% on average) betterImproved
use continual adaptation instead of one-time customization on first 60 s (ablation)AMS · A2D2mIoU: 50.78 -> 69.31Improved
one-time customization versus noneDeepLabV3-MobileNetV2 · A2D2mIoU: 62.05 -> 50.78Hurt
use remote teacher with tracking instead of AMSRemote+Tracking · Outdoor ScenesmIoU: 74.26 -> 69.05Hurt
add gradient-guided selection instead of random selection (1% params) (ablation)AMS coordinate descent · Outdoor ScenesmIoU diff vs full model: -5.29 -> -2.87Improved
select 5% of parameters by gradient guidance rather than updating full model (ablation)AMS · Outdoor ScenesmIoU / downlink bandwidth: -0.73 mIoU; 3302 -> 205 KbpsMixed
update first or last layers only instead of gradient-guided (5%) (ablation)AMS coordinate descent · Outdoor ScenesmIoU diff vs full model: -0.73 -> -8.98 (last), -8.37 (first)Hurt
use longer training horizon (about 4 min) rather than most recent frame (Just-In-Time) (ablation)AMS · Cityscapes, A2D2, Outdoor Scenesdownlink bandwidth for same accuracy: about 10x less bandwidthImproved
remove gradient-guided selection from Just-In-Time (ablation)Just-In-Time · Cityscapes, A2D2, Outdoor Scenesdownlink bandwidth: 150x more bandwidth than AMSHurt
use training horizon 16 s versus longer (64, 256 s) at larger update intervals (ablation)AMS student · Driving in LAaccuracyHurt
add adaptive training rate (ATR) (ablation)AMS multi-client · Outdoor Scenessupported edge devices per V100 with <1% mIoU loss: 7 -> 9Improved
use Just-In-Time default threshold (75%) instead of matched accuracy (ablation)Just-In-Time · All datasetsmIoU / bandwidth: +1.0% mIoU, 3.3x higher bandwidthMixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy