AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Self-Supervised Geometric Features Discovery via Interpretable Attention for Vehicle Re-Identification and Beyond

Ming Li, Xinming Huang, Ziming Zhang

ICCV 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning · 58 citations (OpenAlex)

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 the interpretable-attention geometric branch (GFB, K=7) (ablation)GB baseline · VeRi-776tmAP (%): 84.0 -> 85.9Improved
add the interpretable-attention geometric branch (GFB, K=7) (ablation)GB baseline · CityFlow-ReIDimAP (%): 32.04 -> 36.63Improved
add a plain ResNet18 branch without attention (ablation)GB baseline · CityFlow-ReIDimAP (%): 32.04 -> 34.63Improved
use IAM-based GFB instead of a plain ResNet18 branch (ablation)GB baseline · CityFlow-ReIDimAP (%): 34.63 -> 36.63Improved
add self-supervised rotation learning (SLB, lambda 1.0) to GB+GFB (ablation)GB+GFB · VeRi-776tmAP (%): 85.9 -> 86.2Improved
add self-supervised rotation learning (SLB, lambda 1.0) to GB+GFB (ablation)GB+GFB · CityFlow-ReIDimAP (%): 36.63 -> 37.14Improved
add self-supervised rotation learning (SLB, lambda 1.0) to GB+GFB (ablation)GB+GFB · CityFlow-ReIDTop-1 (%): 59.98 -> 60.08Improved
increase attention neighborhood K from 7 to 15 (ablation)GB+GFB · CityFlow-ReIDimAP (%): 36.63 -> 36.32No clear effect
raise lambda_SLB from 0.1 to 1.0 (ablation)GB+GFB+SLB · VeRi-776tmAP (%): 85.8 -> 86.2No clear effect
raise lambda_SLB from 1.0 to 2.0 (ablation)GB+GFB+SLB · CityFlow-ReIDimAP (%): 37.14 -> 36.54No clear effect
use ours instead of SAVER (no extra supervision)Ours · VehicleID Small galleryTop-1 (%): 79.9 -> 86.8Improved
use ours instead of PVEN (uses extra supervision)Ours · VeRi-776imAP (%): 79.5 -> 81.0Improved

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