ベンチマーク論文,Gait recognition,Shinan zouが第一著者，jin Tangのグループ
Abstract, Introduction
Gait Recognition is the task that analyzes how pedestrians walk from images.
This dataset is more difficult than ever.
The novelty of this method is that covariates exist when analyzing.
Also, parsing is important to increase the accuracy.
Parsing means how the pedestrian is parsed (e.g head, arm, etc)
Related works
Traditionally, poses and silhouettes, and RGB frames are used for features.
Experiment
The proposed method compares how the accuracy changes according to the position of the camera (0, 45, 90,...).
Vaiables
Single covariate
Carrying
Clothing
Road
Speed
Walking styles
Multi covariate
2, 3, 4 mixed covariates exist
感想
面白い論文だと思います．ベンチマーク論文ですが
parsingに注目するのは面白いですが，parsingをすると性能が上がるようにタスク設計がされているようにも感じます
Top comments (0)