AI Paper Atlas

Research map / Detection and segmentation / CNN features

Grid Loss: Detecting Occluded Faces

Michael Opitz, Georg Waltner, Georg Poier, Horst Possegger, Horst Bischof

ECCV 2016 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 76 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 hinge loss with grid loss on output layer2-layer CNN detector · FDDB (284 FP)true positive rate: 0.819 -> 0.851Improved
add grid loss on hidden layers as auxiliary supervision (ablation)grid loss + hinge · FDDB (284 FP)true positive rate: 0.851 -> 0.863Improved
use block size 2 instead of 16 in grid loss (ablation)grid loss + hinge on hidden layers · FDDB (284 FP)true positive rate: 0.834 -> 0.863Improved
decrease lambda toward zero (down-weight part losses) (ablation)grid loss · FDDB (284 FP)true positive rateHurt
use grid loss instead of hinge loss on heavily occluded faces2-layer CNN detector · COFW-HOtrue positive rate: 0.909 -> 0.979Improved
use grid loss instead of hinge loss on less occluded faces2-layer CNN detector · COFW-LOtrue positive rate: 0.982 -> 0.998Improved
train with grid loss instead of hinge loss2-layer CNN detector · Feature map correlationsum of off-diagonal correlation: 22500.25 -> 225.96Improved
train with grid loss on hidden layers when only 1% of positives are used2-layer CNN detector · FDDB (284 FP), 0.01 training fractiontrue positive rate: 0.700 -> 0.802Improved
add hinge loss on hidden layers (deep supervision) to baselinehinge loss CNN · FDDB (284 FP)true positive rate: 0.819 -> 0.834Improved
train ellipse regressor with numerical PASCAL overlap instead of SSE (ablation)ellipse regressor · FDDB continuous evaluation (284 FP)true positive rate: 0.700 -> 0.702No clear effect
add ellipse regressor to detector (ablation)grid loss detector · FDDB (284 FP)true positive rate: 0.863 -> 0.872Improved
skip intermediate pyramid scales with ellipse regressor (ablation)grid loss detector · FDDB (284 FP)true positive rate: 0.872 -> 0.867No clear effect

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