Folder Contents Description
============================

This folder contains two demo videos showcasing inference results from the baseline and the TEMPURA fine-tuned models:

1. Qwen2.5-VL-3B_demo_480p.mp4
   - Inference result from the baseline model (Qwen2.5-VL-3B).

2. TEMPURA_demo_480p.mp4
   - Inference result from our proposed model, TEMPURA fine-tuned from Qwen2.5-VL-3B checkpoint.

Both videos display model-generated timestamps alongside their corresponding event descriptions, rendered as subtitles directly on the video. These demos demonstrate that TEMPURA is capable of producing video-grounded, time-aligned, and detailed event descriptions — outperforming the baseline in generating temporally accurate and semantically rich narrations of video content.
