znhu / hmdb51-action-recognition Goto Github PK
View Code? Open in Web Editor NEWThis project forked from azaddeihim/hmdb51-action-recognition
Created several models to classify actions in HMDB51 dataset. Used models consisting of convolutional and pooling layers for feature extraction followed by recurrent neural networks for classification. Devised novel two-stream architecture capturing spatial and temporal features of input videos, in conjunction with Mask R-CNN for background removal, with bidirectional long-short term memory for classification, earning 80% accuracy on the test set.