NOTE: PyTorch version should >= Pytorch0.4 Installation : https://github.com/mastewalhabtamu/my_detectorn2/blob/master/INSTALL.md Object detection: python detectron2-master/demo/demo_rpn.py --config-file ../configs/COCO-Detection/faster_rcnn_R_101_FPN_3x.yaml --opts MODEL.WEIGHTS detectron2://COCO-Detection/faster_rcnn_R_101_FPN_3x/137851257/model_final_f6e8b1.pkl After the objects are detected, we get the object and pairwise data: python ./gendata/get_object.py python ./gendata/get_pairwise.py Download model： The trained weight and intermediate data has been uploaded to Google Drive，you can download them from '' https://drive.google.com/file/d/11Lv4ROr-OzmCO9cbhLByK5YtzUBBASZI/view?usp=sharing ''. After downloading, copy the 'weights' and 'data' folders to the root directory, and then we will have the structure as: msin/ config data detectron-master feeders gendata graph model weights ensemble.py main.py readme.txt TESTING: To fuse the results of human stream, object stream and pairwise stream, run test firstly to generate the scores of the softmax layer. Considering that the data of human stream (including joints and bones) is large and it is the same with 2s-AGCN, the score of human stream is directly given in this version of our code. The data of human stream can be obtained using the code of 2s-AGCN from this website (https://github.com/lshiwjx/2s-AGCN). C-Setup: python main.py --config ./config/ntu120_setup/test_joint.yaml python main.py --config ./config/ntu120_setup/test_bone.yaml python main.py --config ./config/ntu120_setup/test_object.yaml python main.py --config ./config/ntu120_setup/test_pairwise.yaml C-Subject: python main.py --config ./config/ntu120_subject/test_joint.yaml python main.py --config ./config/ntu120_subject/test_bone.yaml python main.py --config ./config/ntu120_subject/test_object.yaml python main.py --config ./config/ntu120_subject/test_pairwise.yaml Then combine the generated scores with: python ensemble.py Also, we provide the intermediate data (including human skeleton, object appearance, and object position) of the first 6000 video clips in the test set under C-setup protocols making it easy to run our code. Test: python main.py --config ./config/ntu120_setup_tiny/test_joint.yaml python main.py --config ./config/ntu120_setup_tiny/test_bone.yaml python main.py --config ./config/ntu120_setup_tiny/test_object.yaml python main.py --config ./config/ntu120_setup_tiny/test_pairwise.yaml Then combine the generated scores with: python ensemble.py