Interpretable deep learning for human activity recognition (HAR)

Explainable Human Activity Recognition (HAR) system built with PyTorch using a ResNet-50 backbone and BiLSTM-based temporal modeling to classify video actions from the UCF50 dataset. Integrates attention mechanisms and Grad-CAM to provide interpretable, high-accuracy predictions for real-world video analysis applications.