jh502125/triplesumm-demo
0
TripleSumm — AI Video Summarizer
Upload a short video and the model will extract the most important moments by fusing three modalities simultaneously:
- 👁 Visual — CLIP ViT-L/14 frame embeddings
- 🎵 Audio — AST (Audio Spectrogram Transformer) embeddings
- 📝 Speech — Whisper transcription + RoBERTa text embeddings
The TripleSumm model (ICLR 2026) selects up to 15% of the original video using an importance-score-driven knapsack algorithm.
Paper / Code: github.com/smkim37/TripleSumm Lab: AinsteinLab