Olympus provides a universal task router for computer vision, enabling a single model to handle diverse tasks like image generation, 3D model creation, and video synthesis. It is designed for researchers and developers working with multimodal AI systems who need a unified approach to orchestrate complex visual workflows.

How It Works

Olympus acts as a router, interpreting complex, multi-step prompts and directing them to appropriate vision-language models. It leverages a chain-of-action approach, breaking down user requests into sequential sub-tasks. This allows for a more flexible and powerful interaction model, where a single input can trigger a cascade of specialized visual operations.

Quick Start & Requirements

• Install: Clone the repository, create a conda environment (conda create -n olympus python==3.10 -y), activate it (conda activate olympus), and install dependencies (pip install -r requirements.txt).
• Prerequisites: Python 3.10, Conda.
• Models & Data: Download Olympus model (python download_olympus.py), fine-tuning data (python download_olympus_dataset.py), and Mipha-3B model (python download_mipha_3b.py).
• Resources: Requires downloading several model and dataset files.
• Docs: Evaluation.md

Highlighted Details

• CVPR 2025 Highlight paper.
• Supports 20 distinct computer vision tasks.
• Built upon Mipha and LLaVA projects.
• Includes official code, datasets, and models.

Maintenance & Community

• Project is actively maintained with code, training, inference, datasets, and models released.
• Built on Mipha and LLaVA.

Licensing & Compatibility

• License not explicitly stated in the README.
• Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The license is not specified, which may impact commercial adoption. The README does not detail specific hardware requirements beyond standard Python environments.