000 03146nas a2200397 a 4500
005 20260818095404.0
008 240223s2023 th u m tt 000 a eng d
035 _a.b12420591
099 _aAIT Thesis no.CS-23-05
100 1 _aLamsal, Diwas
245 1 0 _aEfficient posevit :
_bexploring efficient vision transformers and training strategies for fast human pose estimation
260 _aPathum Thani, Thailand :
_bAsian Institute of Technology,
_c2023
300 _a74 leaves :
_bill.
490 1 _aThesis ;
_vno.CS-23-05
500 _aA thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Computer Science
502 _aThesis (M. Eng.) - Asian Institute of Technology, 2023
520 _aHuman pose estimation (HPE) serves as a critical component in various downstream applications, including activity recognition and human-computer interaction. However, it faces challenges such as occlusion, complex body poses, diverse clothing and environments, and low-quality images. State-of-the-art HPE methods adept at addressing these challenges often prove to be computationally expensive to run on low-cost edge devices. On the other hand, the efficient models designed to run on edge devices are not as accurate, especially when faced with occlusion. In this thesis, I explored the use of computationally efficient vision transformer (ViT) backbones for the task of HPE to produce an efficient model capable of running on edge devices while also addressing common HPE challenges. Specifically, I replaced the ViT encoder in ViTPose with effi cient ViT variants, and performed a comprehensive comparison. The findings indicate that the EdgeNeXt-Base model offers an optimal speed-accuracy trade-off among the evaluated models. Through knowledge distillation and multi-task training strategies, the resulting models exhibit improved performance, particularly in handling occlusion. Additionally, I produced a wheelchair-based human pose estimation and activity recognition dataset aimed at developing a system that effectively monitors the activities of people sitting in wheelchairs.
650 0 _aImage analysis
650 0 _aHuman-computer interaction
700 1 _aDailey, Matthew N.,
_eChairperson
700 0 _aMongkol Ekpanyapong,
_eExamination Committee
700 0 _aChaklam Silpasuwanchai,
_eExamination Committee
710 2 _aAIT Scholarships,
_eScholarship Donor
810 2 _aAsian Institute of Technology.
_tThesis :
_vno.CS-23-05
856 4 0 _3Full-Text
_uhttp://203.159.5.9/ait-thesis/Viewer/viewer.php?id=B20405
907 _a.b12420591
_bmnait
_ca
902 _a250303
998 _b0
_c240312
_dm
_ea
_fa
_g0
945 _lmnarc
945 _lmnarc
942 _c67
942 _c40
909 _aBarcode : -
_bCREATED : 2024-02-23
_cRECORD # : i13486640
_dLPATRON : 0
_eLCHKIN : -
_f# RENEWALS : 0
_g# OVERDUE : 0
_hIUSE3 : 0
_iTOT CHKOUT : 0
_jTOT RENEW : 0
909 _aBarcode : 30050120900476
_bCREATED : 2025-03-03
_cRECORD # : i13535183
_dLPATRON : 0
_eLCHKIN : -
_f# RENEWALS : 0
_g# OVERDUE : 0
_hIUSE3 : 0
_iTOT CHKOUT : 0
_jTOT RENEW : 0
999 _c35568
_d35568