ucf50-dataset

A Human Action Recognition (HAR) model combining 3D CNN and LSTM networks to accurately recognize actions in videos using spatial-temporal feature extraction. Trained on UCF-50 and outperforming existing architectures.

A combination between ResNet50 and Bi-LSTM, train & test on UCF Dataset to create a Human Activity Recognition (HAR) AI Model.

Developed and implemented deep learning models for human activity recognition, focusing on spatiotemporal data processing and sequence modeling.

Efficient video action recognition using hybrid techniques: combining ORB, SIFT, and deep models like VideoMAE and (2+1)D Conv to reduce data size while maintaining performance.

Robust video action recognition on UCF50 under realistic corruption shifts, with reproducible corruption generation, baseline training/evaluation, ViTTA test-time adaptation, and RMGA adaptation benchmarking.

Cross-modal Human Action Recognition using spatial video representations, temporal sensor signals, and Transformer self-attention.

Recognize human activities in real-time from video

This project evaluates the performance of a Long Short-Term Memory (LSTM) architecture for human movement classification in the UCF50 dataset.