Zero-shot Video Moment Retrieval via Off-the-shelf Multimodal Large Language Models

Name: Zero-shot Video Moment Retrieval via Off-the-shelf Multimodal Large Language Models
Uploaded: 2026-10-08T23:08:21.661177+00:00
Duration: 662 s
Description: Moment-GPT is a tuning-free method for zero-shot Video Moment Retrieval that uses frozen Multimodal Large Language Models (MLLMs). It improves performance by first using LLaMA-3 to correct language biases in search queries. Then, it adaptively generates candidate video spans and selects the best ones using video comprehension modules, achieving state-of-the-art results without any fine-tuning.

Moment-GPT is a tuning-free method for zero-shot Video Moment Retrieval that uses frozen Multimodal Large Language Models (MLLMs). It improves performance by first using LLaMA-3 to correct language biases in search queries. Then, it adaptively generates candidate video spans and selects the best ones using video comprehension modules, achieving state-of-the-art results without any fine-tuning.

Open the full episode