TUTCRIS - Tampereen teknillinen yliopisto

TUTCRIS

Portrait instance segmentation for mobile devices

Tutkimustuotosvertaisarvioitu

Yksityiskohdat

AlkuperäiskieliEnglanti
Otsikko2019 IEEE International Conference on Multimedia and Expo, ICME 2019
KustantajaIEEE
Sivut1630-1635
Sivumäärä6
ISBN (elektroninen)9781538695524
DOI - pysyväislinkit
TilaJulkaistu - 1 heinäkuuta 2019
OKM-julkaisutyyppiA4 Artikkeli konferenssijulkaisussa
TapahtumaIEEE International Conference on Multimedia and Expo - Shanghai, Kiina
Kesto: 8 heinäkuuta 201912 heinäkuuta 2019

Conference

ConferenceIEEE International Conference on Multimedia and Expo
MaaKiina
KaupunkiShanghai
Ajanjakso8/07/1912/07/19

Tiivistelmä

Accurate and efficient portrait instance segmentation has become a crucial enabler for many multimedia applications on mobile devices. We present a novel convolutional neural network (CNN) architecture to explicitly address the long standing problems in portrait segmentation, i.e., semantic coherence and boundary localization. Specifically, we propose a cross-granularity categorical attention mechanism leveraging the deep supervisions to close the semantic gap of CNN feature hierarchy by imposing consistent category-oriented information across layers. Furthermore, a cross-granularity boundary enhancement module is proposed to boost the boundary awareness of deep layers by integrating the shape context cues from shallow layers of the network. We further propose a novel and efficient non-parametric affinity model to achieve efficient instance segmentation on mobile devices. We present a portrait image dataset with instance level annotations dedicated to evaluating portrait instance segmentation algorithms. We evaluate our approach on challenging datasets which obtains state-of-the-art results.

Tutkimusalat

Julkaisufoorumi-taso