ATF: An Alternating Training Framework for Weakly Supervised Face Alignment

Abstract

Together with the impressive progress of algorithms, various benchmark datasets have been released in recent years. Intuitively, it is meaningful to integrate multiple labeled datasets with different annotations to achieve higher performance. Although numerous efforts have been made in joint usage, there yet remain three shortages in recent works, e.g., additional computation, limitation of the markups scheme, and limited support for the regression method. To address the above problems, we proposed a novel Alternating Training Framework (ATF), which leverages similarity and diversity across multi-media sources for a more robust detector. ATF mainly contains two sub-modules: Alternating Training with Decreasing Proportions (ATDP) and Mixed Branch Loss (LMB). In particular, ATDP trains multiple datasets simultaneously via a weakly supervised way to take advantage of the diversity between them, while LMB utilizes similar landmark pairs to constrain different branches of corresponding datasets. And we extend the framework to easily handle three situations: single target detector, joint detector, and novel detector. Extensive experiments show the effectiveness of our framework for both heatmap-based and direct coordinate regression. And we have achieved a joint detector that outperforms SOTA on each benchmark.