AI Paper Atlas

Research map / Detection and segmentation / CNN features

Synergy Between Face Alignment and Tracking via Discriminative Global Consensus Optimization

Muhammad Haris Khan, John McDonagh, Georgios Tzimiropoulos

ICCV 2017 · Detection and segmentation › CNN features · approach: Box regression and loss design · 43 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
combine detection and tracking via ADMM global consensus instead of detection aloneDetection-conv · 300-VW category CAUC@0.08 (%): about 6-7% over the detector across the pt-pt-error spectrumImproved
combine detection and tracking via consensus instead of tracking aloneTracking-conv · 300-VW category CAUC@0.08 (%): 55.94 -> 58.58 (detector); DGCM higher in most videosImproved
use consensus on video #74Detection-conv · 300-VW category C, video #74AUC@0.08 (%): 47.43 -> 51.77Improved
use consensus on video #86Detection-conv · 300-VW category C, video #86AUC@0.08 (%): 62.10 -> 66.13Improved
use consensus on video #99Detection-conv · 300-VW category C, video #99AUC@0.08 (%): 67.88 -> 71.44Improved
use consensus on video #113Detection-conv · 300-VW category C, video #113AUC@0.08 (%): 56.00 -> 58.43Improved
use consensus on video #95 where both detector and tracker work poorlyDetection-conv · 300-VW category C, video #95AUC@0.08 (%): 46.96 -> 44.48Hurt
use consensus on video #100Detection-conv · 300-VW category C, video #100AUC@0.08 (%): 36.98 -> 35.01Hurt
use consensus on video #111Detection-conv · 300-VW category C, video #111AUC@0.08 (%): 63.80 -> 61.38Hurt
use VGG-16 conv-3 features instead of SIFTDGCM · 300-VW category Cfitting accuracyImproved
use more than one detection and one tracking initialization in DGCMDGCM · 300-VWfitting accuracyNo clear effect
use DGCM with 5 initializations instead of PO-CR detector with 5 initializations under high initialization noisePO-CR detector · 300-W test set and cats datasetfitting accuracyImproved

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