Research Intern — Action Recognition (UCF-101; 3D CNN architectures in PyTorch)
Conducted research developing action recognition models using the UCF-101 dataset by evaluating 3D CNN architectures such as I3D and SlowFast. Performed analysis of existing state-of-the-art methods to design and refine a novel model architecture. Implemented experiments and learning iterations in PyTorch to improve action recognition performance. • Worked with video-based action recognition data pipelines. • Compared model families (I3D vs SlowFast) for feature extraction. • Analyzed prior approaches to guide architectural improvements. • Built and trained models using deep learning workflows.