---
library_name: transformers
pipeline_tag: video-text-to-text
license: apache-2.0
---

<div align='center'>
<h1>Video-Panda: Parameter-efficient Alignment for Encoder-free Video-Language Models</h1h1>

[Jinhui Yi*](https://scholar.google.com/citations?user=kLZxzzUAAAAJ&hl=en),
[Syed Talal Wasim*](https://talalwasim.github.io),
[Yanan Luo*](https://scholar.google.com/citations?user=yuDQY0YAAAAJ&hl=en),
[Muzammal Naseer](https://muzammal-naseer.netlify.app/),
[Juergen Gall](https://pages.iai.uni-bonn.de/gall_juergen/)

*Equal Contribution

University of Bonn; Lamarr Institute for Machine Learning and Artificial Intelligence; Khalifa University
<!-- <sup>1</sup> [University of Bonn], <sup>2</sup> [Lamarr Institute for Machine Learning and Artificial Intelligence], <sup>3</sup> [Khalifa University]<br><sup>*</sup> Equal Contribution -->

|  [Paper](https://arxiv.org/abs/2412.18609) | [Code](https://github.com/jh-yi/Video-Panda) |
</div>
We present an efficient encoder-free approach for video-language understanding that achieves competitive performance while significantly reducing computational overhead. Current video-language models typically rely on heavyweight image encoders (300M-1.1B parameters) or video encoders (1B-1.4B parameters), creating a substantial computational burden when processing multi-frame videos. Our method introduces a novel Spatio-Temporal Alignment Block (STAB) that directly processes video inputs without requiring pre-trained encoders while using only 45M parameters for visual processing - at least a 6.5x reduction compared to traditional approaches. The STAB architecture combines Local Spatio-Temporal Encoding for fine-grained feature extraction, efficient spatial downsampling through learned attention and separate mechanisms for modeling frame-level and video-level relationships. Our model achieves comparable or superior performance to encoder-based approaches for open-ended video question answering on standard benchmarks. The fine-grained video question-answering evaluation demonstrates our model's effectiveness, outperforming the encoder-based approaches Video-ChatGPT and Video-LLaVA in key aspects like correctness and temporal understanding. Extensive ablation studies validate our architectural choices and demonstrate the effectiveness of our spatio-temporal modeling approach while achieving 3-4x faster processing speeds than previous methods.

## Model Weights
We release the pretrained and instruction-tuned weights of **Video-Panda** in this repository. 

## ✒️ Citation
If **Video-Panda** is helpful for your research, please consider **star** ⭐ and **citation** 📝 :
```bibtex
@article{yi2024video-panda,
    author    = {Jinhui Yi* and Syed Talal Wasim* and Yanan Luo* and Muzammal Naseer and Juergen Gall},
    title     = {Video-Panda: Parameter-efficient Alignment for Encoder-free Video-Language Models},
    journal   = {arXiv preprint, arXiv:2412.18609},
    year      = {2024},
}
```