An end-to-end video summarization pipeline implementing and evaluating Compact BiLSTM and Interpretable Transformer architectures for keyframe selection. Features frozen MobileNetV3 feature extraction, custom Top-K duration budget policies, and performance benchmarking on the TVSum and SumMe datasets.
computer-vision deep-learning transformers video-summarization pytorch lstm attention-mechanism sequence-modeling keyframe-extraction summe tvsum
-
Updated
Jan 1, 2026 - Jupyter Notebook