This repo contains the reference source code for the paper--- Beyond RNNs: Positional Self-Attention with Co-Attention for Video Question
We have added a visual interface for testing results.
ResNet feature address: https://drive.google.com/open?id=0B15H16jpV4w2SlVleTBRT3dUTGs
C3D feature: https://drive.google.com/open?id=0B15H16jpV4w2cFZoOXpPMlFLX3M
The model weght we trained:https://drive.google.com/drive/folders/1Sg9Q-emVwFq5HOWXYBDilLPuBMHy6MwQ?usp=sharing
You can test by running main_test_Action.py or main_test_frameqa.py in \project_vqa\PSAC-master\code_file
You can train model by running train_Action.py or train_frameqa.py in \project_vqa\PSAC-master\code_file.