Skip to content

NguyenThu512/Action-Identification-with-Fusion-of-BERT-and-3DCNN-for-Smart-Home-Systems

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 

Repository files navigation

Action-Identification-with-Fusion-of-BERT-and-3DCNN-for-Smart-Home-Systems

Key features

  • A fusion of SlowFast and Bidirectional Encoder Representation from Transformer (BERT) for action identification.
  • Efficently exposing temporal relationship between video frames using BERT's Attention Mechanism.
  • Two modalities in fusing SlowFast and BERT for action identification: late-ensemble and early-ensemble.
  • Kinetics51 dataset: a derived dataset from Kinetics400.

Late-ensemble SlowFast-BERT

Early-ensemble SlowFast-BERT

Datasets:

Benchmark

Experiments on HMDB51 dataset

Model Dimensionality reduction Combine BERT Pretrained dataset Top-1 accuracy Top-3 accuracy
SlowFast No No Kinetics400 77.53% 90.87%
SlowFast FRAB BERT early FRAB Early-ensemble Kinetics400 76.62% 91.03%
SlowFast FRAB BERT late FRAB Late-ensemble Kinetics400 78.05% 91.63%
SlowFast FRMB BERT early FRMB Early-ensemble No 76.40% 91.40%
SlowFast FRMB BERT late FRMB Late-ensemble No 78.73% 91.40%

Comparison with state-of-the-art on HMDB51 dataset

Model Pretrained dataset Year Top-1 accuracy
Shuffle and Learn (AlexNet) UCF101 Mar 2016 19.8%
OPN (VGG-M-2048) UCF101 Oct 2017 23.8%
3D Cubic Puzzles (3D ResNet-18) Kinetics400 Nov 2018 33.7%
DPC (Modified 3D Resnet-34) Kinetics400 Sep 2019 35.7%
XDC IG-Kinetics Dec 2020 68.9%
CVRL (R3D-152 2x; K600) Kinetics600 Jun 2021 69.9%
BraVe:V-FA (TSM-50x2) Oct 2021 70.5%
pBYOL Kinetics400 Nov 2021 75.0%
SlowFast FRAB BERT late (Ours) Kinetics400 Dec 2022 78.05%

Experiments on Kinetics51 dataset

Model Dimensionality reduction Combine BERT Pretrained dataset Top-1 accuracy Top-3 accuracy
SlowFast No No Kinetics400 79.08% 93.14%
SlowFast FRAB BERT late FRAB Late-ensemble Kinetics400 82.52% 95.02%
SlowFast FRAB BERT early FRAB Early-ensemble Kinetics400 82.43% 94.61%

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published