Child action and gender recognition based on dataset provided by Prof. Sandygulova. The models are created and pretrained by @open-mmlab and @microsoft
Confusion matrix with the CSN model, optical flow (97.62% accuracy):
Confusion matrix with the i3D model, rgb frames (96.83% accuracy):
Confusion matrix with the TPN model, rgb frames (96.83% accuracy):
Confusion matrix with the TANnet model, optical flow (96.03% accuracy):
Confusion matrix with the TSN model, video (94.44% accuracy):
Confusion matrix with the TSM model, video (93.65% accuracy):
Confusion matrix with the R2+1D model, optical flow (92.68% accuracy):
Confusion matrix with the SlowFast model, video (91.27% accuracy):
Confusion matrix with the TIN model, optical flow (87.30% accuracy):