Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Binarized Convolutional Landmark Localizers for Human Pose Estimation and Face Alignment with Limited Resources

About

Our goal is to design architectures that retain the groundbreaking performance of CNNs for landmark localization and at the same time are lightweight, compact and suitable for applications with limited computational resources. To this end, we make the following contributions: (a) we are the first to study the effect of neural network binarization on localization tasks, namely human pose estimation and face alignment. We exhaustively evaluate various design choices, identify performance bottlenecks, and more importantly propose multiple orthogonal ways to boost performance. (b) Based on our analysis, we propose a novel hierarchical, parallel and multi-scale residual architecture that yields large performance improvement over the standard bottleneck block while having the same number of parameters, thus bridging the gap between the original network and its binarized counterpart. (c) We perform a large number of ablation studies that shed light on the properties and the performance of the proposed block. (d) We present results for experiments on the most challenging datasets for human pose estimation and face alignment, reporting in many cases state-of-the-art performance. Code can be downloaded from https://www.adrianbulat.com/binary-cnn-landmarks

Adrian Bulat, Georgios Tzimiropoulos• 2017

Related benchmarks

TaskDatasetResultRank
Face AlignmentAFLW 2000-3D 68 pts (test)
Mean NME3.26
82
2D Human Pose EstimationMPII (val)
Head97.4
61
Face AlignmentAFLW 21 landmarks
NME2.85
37
Face AlignmentAFLW2000-3D (test)
NME (Full height)3.26
29
Face AlignmentAFLW--
12
Face AlignmentAFLW PIFA protocol (test)
NME4.47
11
3D Face AlignmentAFLW2000-3D--
11
Face AlignmentAFLW-PIFA visible landmarks only
NME0.0302
10
Face AlignmentAFLW (test)
NME ([0,30])2.77
6
Showing 9 of 9 rows

Other info

Code

Follow for update