[ECCV2024] VisionLLaMA: A Unified LLaMA Backbone for Vision Tasks

Introduction

Large language models are built on top of a transformer-based architecture to process textual inputs. For example, the LLaMA stands out among many open-source implementations. Can the same transformer be used to process 2D images? In this paper, we answer this question by unveiling a LLaMA-like vision transformer in plain and pyramid forms, termed VisionLLaMA, which is tailored for this purpose. VisionLLaMA is a unified and generic modelling framework for solving most vision tasks. We extensively evaluate its effectiveness using typical pre-training paradigms in a good portion of downstream tasks of image perception and especially image generation. In many cases, VisionLLaMA have exhibited substantial gains over the previous state-of-the-art vision transformers. We believe that VisionLLaMA can serve as a strong new baseline model for vision generation and understanding.

Generation

DITLLaMA

Please refer to DiTLLaMA.md

SITLLaMA

Please refer to SiTLLaMA.md

UnderStanding

Pretrain using MIM

The pre-training instruction is in PRETRAIN.md.

ImageNet 1k Supervised Training

Please refer to ImageNet1k_SFT

ADE 20k Segmentation

Please refer to Segmentation.md.

COCO Detection

Please refer to Detection.md.

✏️ Reference

If you find VisionLLaMA useful in your research or applications, please consider giving a star ⭐ and citing using the following BibTeX:

@inproceedings{chu2024visionllama,
  title={VisionLLaMA: A Unified LLaMA Backbone for Vision Tasks},
  author={Chu, Xiangxiang and Su, Jianlin and Zhang, Bo and Shen, Chunhua},
  booktitle={European Conference on Computer Vision},
  year={2024}
}