FreeEnricher: Enriching Face Landmarks without Additional Cost

Recent years have witnessed significant growth of face alignment. Though dense facial landmark is highly demanded in various scenarios, e.g., cosmetic medicine and beautification, most works only consider sparse To address this problem, we present a framework that can enrich density by existing datasets, 300W with 68 points WFLW 98 points. Firstly, observe the local patches along each semantic contour are similar appearance. Then, propose weakly-supervised idea learning refinement ability on original landmarks adapting to enriched landmarks. Meanwhile, several operators devised organized together implement idea. Finally, trained model applied as plug-and-play module alignment networks. evaluate our method, manually label testset. Our method yields state-of-the-art accuracy not newly-constructed testset but also testsets without additional cost.