AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

Learning Distilled Collaboration Graph for Multi-Agent Perception

Yiming Li, Shunli Ren, Pengxiang Wu, Siheng Chen, Chen Feng, Wenjun Zhang

NeurIPS 2021 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Collaborative multi-agent perception · 17 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 knowledge distillation on post-collaboration feature maps from early-collaboration teacher (ablation)DiscoGraph · V2X-Sim 1.0AP@IoU=0.5: 57.2 -> 60.3Improved
add knowledge distillation (ablation)DiscoGraph · V2X-Sim 1.0AP@IoU=0.7: 52.3 -> 53.9Improved
add KD to Max aggregation (ablation)Max · V2X-Sim 1.0AP@IoU=0.5: 56.7 -> 56.7No clear effect
add KD to Sum aggregation (ablation)Sum · V2X-Sim 1.0AP@IoU=0.7: 50.9 -> 46.3Hurt
add KD to Cat aggregation (ablation)Cat · V2X-Sim 1.0AP@IoU=0.5: 55.0 -> 57.5Improved
use matrix-valued edge weight instead of scalar weighted average (both with KD) (ablation)DiscoGraph · V2X-Sim 1.0AP@IoU=0.5: 56.7 -> 60.3Improved
use matrix-valued edge weight instead of scalar weighted average (no KD) (ablation)DiscoGraph · V2X-Sim 1.0AP@IoU=0.5: 56.1 -> 57.2Improved
apply KD on more decoder layers: M{1,2,3} vs only H (ablation)DiscoNet · V2X-Sim 1.0AP@IoU=0.5: 57.5 -> 60.3Improved
add KD on fourth decoder layer M{4} (ablation)DiscoNet · V2X-Sim 1.0AP@IoU=0.5: 60.3 -> 59.1Hurt
drop KD on H while keeping KD on M{1-4} (ablation)DiscoNet · V2X-Sim 1.0AP@IoU=0.5: 59.1 -> 58.3Hurt
apply late collaboration to upper boundUpper-bound · V2X-Sim 1.0AP@IoU=0.5: 63.3 -> 59.7Hurt
apply late collaboration to lower boundLower-bound · V2X-Sim 1.0AP@IoU=0.5: 45.8 -> 57.6Improved

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