This repository contains all the code for the paper: "Enhancing Video Vision-Language Models for Camera and Scene Understanding" written for the Deep Learning course at Cornell Tech, Fall 2025.
Run the notebooks in caption-generation-notebooks/:
Run the notebooks in the evaluation-notebooks/:
All prompt templates used in the experiments (both unified dual-task and separate camera/scene prompts) are defined in the paper’s Appendix for reference.
We use the CameraBench dataset:
- CameraBench test split (N ≈ 1,071 videos) for evaluation.
- Please refer to the official CameraBench repository for data access and licensing details.
This repository does not redistribute the dataset. You must download it separately and update the data paths in the notebooks.
Parts of this repository were adapted from:
- Official CameraBench code and examples (data loading, evaluation utilities).
- Qwen2.5-VL-7B example scripts.
- The use of AI to assist in building out the caption generation and evaluation pipelines.
We acknowledge these sources in accordance with the course instructions.
If you refer to this project, please cite the course report:
S. Dong, W. Li, A. Tjokro. “Enhancing Video Vision-Language Models for Camera and Scene Understanding.” Deep Learning, Cornell Tech, 2025.