llava_uground

• 🏠Homepage
• 📖Paper
• 😊Model Weights
• 😊Live Demo (Try it out yourself!)
• Main Code Repo

Note: This is the codebase to inference and train UGround-v1. It is modified from haotian-liu/LLaVA, with changes mainly on image processing. We may use other codebases or architectures for later versions.

Install

1. Create a python environment

conda create -n llava_uground python=3.11 -y
conda activate llava_uground
pip install --upgrade pip  # enable PEP 660 support

2. Install the dependencies

There are several ways to install the package:

# Install from Github
pip install git+https://github.com/boyugou/llava_uground.git

# Install locally
git clone https://github.com/boyugou/llava_uground.git
pip install -e .

# Install from pypi (Not uploaded yet)
pip install to_be_uploaded

Inference

python single_infer.py # single request

python batch_infer.py # for inference of a jsonl