AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Trunk-Branch Ensemble Convolutional Neural Networks for Video-Based Face Recognition

Changxing Ding, Dacheng Tao

TPAMI 2017 · Label-efficient and robust learning › Recognition and long-tail · approach: Margin-based face recognition losses · 440 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 simulated blurred video frames to still-image training data (TS vs SI) (ablation)TBE-CNN trunk · PaSC handheld setverification rate at 1% FAR (%): 68.20 -> 79.33Improved
add simulated blurred video frames to still-image training data (TS vs SI) (ablation)TBE-CNN trunk · PaSC control setverification rate at 1% FAR (%): 83.72 -> 87.31Improved
use simulated blurred video (SV) instead of still images (SI) (ablation)trunk CNN · PaSC handheld setverification rate at 1% FAR (%): 68.20 -> 73.80Improved
use simulated blurred video instead of still images (ablation)trunk CNN · PaSC control setverification rate at 1% FAR (%): 83.72 -> 81.89Hurt
train from scratch on TS data instead of fine-tuning with real video (FT) (ablation)trunk CNN · PaSC handheld setverification rate at 1% FAR (%): 73.70 -> 79.33Improved
add mean-distance regularisation (MDR-TL instead of triplet loss) (ablation)TBE-CNN · PaSC handheld set (SI data, semi-hard)verification rate at 1% FAR (%): 87.00 -> 89.60Improved
add mean-distance regularisation (MDR-TL instead of triplet loss) (ablation)TBE-CNN · PaSC handheld set (TS data, semi-hard)verification rate at 1% FAR (%): 92.46 -> 93.91Improved
add mean-distance regularisation (MDR-TL instead of triplet loss) (ablation)TBE-CNN · PaSC handheld set (TS data, hardest)verification rate at 1% FAR (%): 91.47 -> 93.12Improved
use hardest negative mining instead of semi-hard (ablation)triplet loss / MDR-TL · PaSCverification rate at 1% FAR (%): 92.46 -> 91.47Hurt
use hard instead of semi-hard negative mining (ablation)triplet loss · PaSC handheld set (TS data)verification rate at 1% FAR (%): 92.46 -> 92.73No clear effect
add trunk-branch fusion of facial component patches (TBE-CNN vs trunk network) without BN (ablation)TBE-CNN · PaSC handheld setverification rate at 1% FAR (%): 5.82% marginImproved
add trunk-branch fusion of facial component patches with BN layers (ablation)TBE-CNN · PaSC handheld setverification rate at 1% FAR (%): 3.48% marginImproved

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