AI Paper Atlas

Research map / Detection and segmentation / CNN features

GET: Group Event Transformer for Event-Based Vision

Yansong Peng, Yueyi Zhang, Zhiwei Xiong, Xiaoyan Sun, Feng Wu

ICCV 2023 · Detection and segmentation › CNN features · approach: Spiking event-camera detection · 81 citations (OpenAlex)

This paper tried a common change: Add attention module

add Event Dual Self-Attention block · Cifar10-DVS · top-1 accuracy (%): 79.9 -> 81.5 Improved

What 492 accepted papers report for this change

See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.

Read the paper

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 Group Token Embedding in place of voxel-grid patch embedding (ablation)GET (Nested-T backbone) · Cifar10-DVStop-1 accuracy (%): 77.2 -> 79.9Improved
add Group Token Embedding in place of voxel-grid patch embedding (ablation)GET (Nested-T backbone) · 1MpxmAP (%): 37.5 -> 39.1; runtime 32.1 -> 16.2 msImproved
add Event Dual Self-Attention block (ablation)GET · Cifar10-DVStop-1 accuracy (%): 79.9 -> 81.5Improved
add Group Token Aggregation without EDSA (ablation)GET (Model II) · Cifar10-DVStop-1 accuracy (%): 79.9 -> 79.4Hurt
add GTA together with EDSA (full GET) (ablation)GET · Cifar10-DVStop-1 accuracy (%): 81.5 -> 84.8Improved
replace Group Token with other event representations (ablation)GET · Cifar10-DVStop-1 accuracy (%): 84.8 vs 80.4 (TORE best alternative)Improved
use dual residual EDSA block instead of serial or parallel SSA+GSA (ablation)GET · Cifar10-DVStop-1 accuracy (%): 81.9 -> 84.8Improved
use naive parallel SSA+GSA block instead of SSA only (ablation)GET · Cifar10-DVStop-1 accuracy (%): 79.4 -> 78.5Hurt
increase group number G beyond 12 on short-stream data (ablation)GET · Cifar10-DVStop-1 accuracy (%): 84.8 -> 82.3 -> 77.6 (G=12, 24, 48)Hurt
use larger group number G=24 on long-stream data (ablation)GET · DVS128Gesturetop-1 accuracy (%): 97.2 -> 97.9Improved
replace Swin-T v2/Nested-T modules with GET modules (token embedding, attention, aggregation) (ablation)GET · Cifar10-DVStop-1 accuracy (%): 80.4/81.0 -> 84.8 (aggregation)Improved
add ConvLSTM memory to GETGET · Gen1mAP (%): 38.7 -> 47.9Improved

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