List your company

Video Summarization on MovieSum (test)

34.47Visual Reconstruction ScoreGold Screenplay
13.61819.031524.44529.8585Dec 12, 2025

Evaluation results

MethodLinks
Gold ScreenplayInput Source=Gold Screenplay, Summarization model=Gemini 2.5 Flash2025.1234.4735.4834.9747.4311.8845.3434.06
Built Screenplay (reference clips)Clip selection strategy=reference clips, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1222.4335.3828.947.2811.6745.2134.14
our clips two-shot (Qwen2.5-Omni-7B)Number of clips (K)=75, Clip selection strategy=our clips, Evaluation protocol=two-shot, Selection model=Qwen2.5-Omni-7B, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1222.2533.4527.8546.5310.844.3233.55
our clips two-shot (Qwen2.5-Omni-7B)Number of clips (K)=25, Clip selection strategy=our clips, Evaluation protocol=two-shot, Selection model=Qwen2.5-Omni-7B, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1221.0534.0127.5346.911.5544.6933.93
our clips zero-shot (Qwen2.5-Omni-7B)Number of clips (K)=50, Clip selection strategy=our clips, Evaluation protocol=zero-shot, Selection model=Qwen2.5-Omni-7B, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1220.9733.6327.346.5411.2244.4933.63
our clips zero-shot (Qwen2.5-Omni-7B)Number of clips (K)=75, Clip selection strategy=our clips, Evaluation protocol=zero-shot, Selection model=Qwen2.5-Omni-7B, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1220.8731.2526.0646.0410.7543.8733.13
our clips zero-shot (Qwen2.5-Omni-7B)Number of clips (K)=25, Clip selection strategy=our clips, Evaluation protocol=zero-shot, Selection model=Qwen2.5-Omni-7B, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1220.8135.6428.2346.2911.244.1633.59
our clips two-shot (Qwen2.5-Omni-7B)Number of clips (K)=50, Clip selection strategy=our clips, Evaluation protocol=two-shot, Selection model=Qwen2.5-Omni-7B, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1220.634.2227.4146.2210.9744.133.46
silent clipsNumber of clips (K)=75, Clip selection strategy=silent clips, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1219.4332.4225.924610.8443.8333.37
silent clipsNumber of clips (K)=50, Clip selection strategy=silent clips, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1217.5331.8224.6746.111.3243.9533.36
silent clipsNumber of clips (K)=25, Clip selection strategy=silent clips, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1216.3332.6124.4746.2410.9544.1133.24
random clipsNumber of clips (K)=50, Clip selection strategy=random clips, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1215.6830.9123.2946.1110.5143.8833.21
random clipsNumber of clips (K)=25, Clip selection strategy=random clips, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1215.4431.9323.6946.1110.6144.0733.27
random clipsNumber of clips (K)=75, Clip selection strategy=random clips, Recaptioning model=Gemini 2.5 Flash-Lite, Summarization model=Gemini 2.5 Flash2025.1214.6830.922.7945.3510.3443.1532.73
Transcripts (no video)Input Source=Transcripts, Summarization model=Gemini 2.5 Flash2025.1214.4226.8920.6544.6610.3542.6432.12