Please enable JavaScript.
Coggle requires JavaScript to display documents.
Child Action Research Paper - Coggle Diagram
Child Action Research Paper
contributions
First Implementation of GCN based methods on child actions :check:
Comparison of results between protocols with acceptable accuracies :check:
Pre trained model's dataset classes affect the feature representation :check:
Small dataset results encourages further child action implementations with large datasets :check:
results based on experiments? :!: [acceptable accuracy despite skeleton struc. change ; important for robust skeleton based models :check: ]
Protocols
Child Full Body Gesture
Dissimilar
Full Dataset
Similar
Kinder-Gater
Similar
Full Dataset
Dissimilar
Random
KG on Trained CFBG :!:
Pre-Trained Models
NTU - 60
NTU 120
NTU 44
NTU 51
NTU 22
Learning Methods
Transfer Learning
Feature Extraction
Curriculum Learning
From Scratch
Experiments
child vs adult :!:
skeleton structure
frame rate effect
Pre trained model
size
content
skeleton size effect :!:
Comparision
ST-GCN
2s-AGCN :warning:
TCN :!!:
DGNN :red_cross:
Discussion
based on confusion matrix results
based on pre trained model accuracies