This repository contains the code for our CVPR 2025 Workshop on Video LLMs submission, "How Important are Videos for Training Video LLMs?. It provides a simple framework for constructing pseudovideo question-answer samples from datasets of annotated images. It also provides a training pipeline based on the LongVU work, with inference on the TVBench dataset.
VisualComputingInstitute/videollm-pseudovideo-training
Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|