AI Paper Atlas

Research map / Graph learning / Knowledge and temporal graphs

Graph-Based Asynchronous Event Processing for Rapid Object Recognition

Yijin Li, Han Zhou, Bangbang Yang, Ye Zhang, Zhaopeng Cui, Hujun Bao, Guofeng Zhang

ICCV 2021 · Graph learning › Knowledge and temporal graphs · approach: Temporal graph neural architectures · 117 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
replace graph convolution with slide convolution (ablation)EvS · N-Caltech101computation per event (Mps/ev): 1152 -> 11.5Improved
replace graph convolution with slide convolution (ablation)EvS · CIFAR10-DVScomputation per event (Mps/ev): 3020 -> 33.2Improved
replace graph convolution with slide convolution (ablation)EvS · N-Caltech101accuracy: 0.761 -> 0.761No clear effect
replace graph convolution with slide convolution (ablation)NVS · N-Caltech101computation per event (Mps/ev): 221 -> 7.8Improved
replace graph convolution with slide convolution (ablation)EvS · N-Caltech101processing time per event update: 130.4 ms -> 16.9 msImproved
replace k-d tree (nanoflann) with pixel-queue radius search (ablation)graph construction · synthetic window of 100,000 eventssearch time: reduces the time cost by halfImproved
replace k-d tree (nanoflann) with pixel-queue radius search (ablation)graph construction · synthetic window of 100,000 eventsinsertion and deletion cost: an order of magnitude lowerImproved
use event-wise SlideGCN instead of batch-wise 50 ms processingEvS · N-Caltech101 (stream)accuracy: 0.761 -> 0.669Hurt
use event-wise SlideGCN instead of batch-wise 50 ms processingEvS · N-Caltech101 (stream)event index at response: 19154 -> 3118Improved
reduce batch size from 50 ms to 10 ms (ablation)batch-wise EvS · N-Caltech101 (stream)accuracy: 0.761 -> 0.362Hurt
increase mini-batch size from 1 to 100 (ablation)SlideGCN · N-Caltech101 (stream)cumulative MFLOPS: 17711 -> 3041Improved
increase mini-batch size from 1 to 100 (ablation)SlideGCN · N-Caltech101 (stream)average MFLOPS: 5.68 -> 95.64Hurt

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