R21D_basic_PyTorch

still in development...

This is the repository of my R(2+1)D model based on ResNet-18. Development architecture is PyTorch 1.0.

Details about R(2+1)D can be referred from A Closer Look at Spatiotemporal Convolutions for Action Recognition (Du Tran .etc)

cd data/kinetics400
mkdir access && cd access
ln -s $YOUR_KINETICS400_DATASET_TRAIN_DIR$ RGB_train
ln -s $YOUR_KINETICS400_DATASET_VAL_DIR$ RGB_val

Executing shell script in the scripts directory:

Train: ./scripts/train_kinetics400_21d.sh

Test: ./scripts/test_kinetics400_21d.sh

Training and testing environment: 8 GTX 1080 Ti GPUs

Input sample size: 16 * 3 * 16 * 112 * 112 on each GPU

Results:

Thanks to Dr. Wang's guiding and Lei Zhou's help.