A PyTorch implementation of R2Plus1D and C3D based on CVPR 2017 paper "A Closer Look at Spatiotemporal Convolutions for Action Recognition" and CVPR 2014 paper "Learning Spatiotemporal Features with 3D Convolutional Networks"
58
stars
14
forks
source link
I train R2Plus1D on the UCF101(I only select 51 classes) from scratch. The val acc is only 40%. #5
I train it on a sigle gpu(v100), and I don't adjust the parameter settings in the code.