| <div align="center"> |
|
|
| <h2><a href="https://arxiv.org/abs/2408.10605">MUSES: 3D-Controllable Image Generation via Multi-Modal Agent Collaboration</a></h2> |
|
|
|
|
| [Yanbo Ding*](https://github.com/DINGYANB), |
| [Shaobin Zhuang](https://scholar.google.com/citations?user=PGaDirMAAAAJ&hl=zh-CN&oi=ao), |
| [Kunchang Li](https://scholar.google.com/citations?user=D4tLSbsAAAAJ), |
| [Zhengrong Yue](https://arxiv.org/search/?searchtype=author&query=Zhengrong%20Yue), |
| [Yu Qiao](https://scholar.google.com/citations?user=gFtI-8QAAAAJ&hl), |
| [Yali Wang†](https://scholar.google.com/citations?user=hD948dkAAAAJ) |
|
|
| [](https://arxiv.org/abs/2408.10605) [](https://github.com/DINGYANB/MUSES) [](https://huggingface.co/yanboding/MUSES/) |
|
|
| </div> |
|
|
| ## 💡 Motivation |
|
|
| Despite recent advancements in text-to-image generation, most existing methods struggle to create images with multiple objects and complex spatial relationships in 3D world. To tackle this limitation, we introduce a generic AI system, namely MUSES, for 3D-controllable image generation from user queries. |
|
|
| <img alt="image" src="https://huggingface.co/yanboding/MUSES/resolve/main/demo.png"> |
| </a> |
|
|
|
|
| ## 🤖 Architecture |
|
|
| Our MUSES realize 3D controllable image generation by developing a progressive workflow with three key components, including: |
| 1. Layout Manager for 2D-to-3D layout lifting; |
| 2. Model Engineer for 3D object acquisition and calibration; |
| 3. Image Artist for 3D-to-2D image rendering. |
|
|
| By mimicking the collaboration of human professionals, this multi-modal agent pipeline facilitates the effective and automatic creation of images with 3D-controllable objects, through an explainable integration of top-down planning and bottom-up generation. |
|
|
| <img alt="image" src="https://huggingface.co/yanboding/MUSES/resolve/main/overview.png"> |
| </a> |
|
|
|
|
| ## 🔨 Installation |
|
|
| 1. Clone this GitHub repository and install the required packages: |
|
|
| ``` shell |
| git clone https://github.com/DINGYANB/MUSES.git |
| cd MUSES |
| |
| conda create -n MUSES python=3.10 |
| conda activate MUSES |
| |
| pip install -r requirements.txt |
| ``` |
| |
|
|
| 2. Download other required models: |
|
|
| | Model | Storage Path | Role | |
| |----------------------|----------------------|-------------| |
| | [OpenAI ViT-L-14](https://huggingface.co/openai/clip-vit-large-patch14) | `model/CLIP/` | Similarity Comparison | |
| | [Meta Llama-3-8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B) | `model/Llama3/` | 3D Layout Planning | |
| | [stabilityai stable-diffusion-3-medium (SD3)](https://huggingface.co/stabilityai/stable-diffusion-3-medium) | `model/SD3-Base/` | Image Generation | |
| | [InstantX SD3-Canny-ControlNet](https://huggingface.co/InstantX/SD3-Controlnet-Canny) | `model/SD3-ControlNet-Canny/` | Controllable Image Generation | |
| | [examples_features.npy](https://huggingface.co/yanboding/MUSES/upload/main) | `/dataset/` | In-Context Learning | |
| | [finetuned_clip_epoch_20.pth](https://huggingface.co/yanboding/MUSES/upload/main) | `/model/CLIP/` | Orientation Calibration | |
| |
| Since our MUSES is a training-free multi-model collaboration system, feel free to replace the generative models with other competitive ones. For example, we recommend users to replace the Llama-3-8B with more powerful LLMs like [Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B) and [GPT 4o](https://platform.openai.com/docs/models/gpt-4o). |
| |
| 3. *Optional* Downloads: |
| - Download our self-built 3D model shop at this [link](https://huggingface.co/yanboding/MUSES/upload/main), which includes 300 high-quality 3D models, and 1500 images of various objects with different orientations for fine-tuing the [CLIP](https://huggingface.co/openai/clip-vit-base-patch32). |
| - Download multiple ControlNets such as [SD3-Tile-ControlNet](https://huggingface.co/InstantX/SD3-Controlnet-Tile), [SDXL-Canny-ControlNet](https://huggingface.co/TheMistoAI/MistoLine), [SDXL-Depth-ControlNet](https://huggingface.co/diffusers/controlnet-zoe-depth-sdxl-1.0), and other image generation models, e.g., [SDXL](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) with [VAE](https://huggingface.co/madebyollin/sdxl-vae-fp16-fix). |
|
|
|
|
| ## 🌟 Usage |
| Use the following command to generate images. |
| ``` shell |
| cd MUSES && bash multi_runs.sh "test_prompt.txt" "test" |
| ``` |
| Where the **first argument** is the input txt file containing the prompts in rows, and the **second argument** is the identifier of the current run, which will be appended to the output folder name. For SD3-Canny-ControlNet, each prompt results in 5 images of different control scales. |
|
|
|
|
| ## 📊 Dataset & Benchmark |
| ### Expanded NSR-1K |
| Since the original [NSR-1K](https://github.com/Karine-Huang/T2I-CompBench) dataset lacks layouts in 3D scenes and complex scenes, so we manually add some |
| prompts with corresponding layouts. |
| Our expanded NSR-1K dataset is in the directory `dataset/NSR-1K-Expand`. |
|
|
| ### Benchmark Evaluation |
|
|
| For *T2I-CompBench* evaluation, we follow its official evaluation codes in this [link](https://github.com/Karine-Huang/T2I-CompBench). Note that we choose the best score among the 5 images as the final score. |
|
|
| Since T2I-CompBench lacks detailed descriptions of complex 3D spatial relationships of multiple objects, we construct our T2I-3DisBench (`dataset/T2I-3DisBench.txt`), which describes diverse 3D image scenes with 50 detailed prompts. |
| For *T2I-3DisBench* evaluation, we employ [Mini-InternVL-2B-1.5](https://huggingface.co/OpenGVLab/Mini-InternVL-Chat-2B-V1-5) to score the generated images from 0.0 to 1.0 across four dimensions, including object count, object orientation, 3D spatial relationship, and camera view. You can download the weights at this [link](https://huggingface.co/OpenGVLab/Mini-InternVL-Chat-2B-V1-5) and put them into the folder `model/InternVL/`. |
|
|
| ``` shell |
| python inference_code/internvl_vqa.py |
| ``` |
|
|
| After running it, it will output an average score. |
| Our MUSES demonstrates state-of-the-art performance on both benchmarks, verifying its effectiveness. |
|
|
| ## 💙 Acknowledgement |
| MUSES is built upon |
| [Llama](https://github.com/meta-llama/llama3), |
| [NSR-1K](https://github.com/Karine-Huang/T2I-CompBench), |
| [Shap-e](https://github.com/openai/shap-e), |
| [CLIP](https://github.com/openai/CLIP), |
| [SD](https://github.com/Stability-AI/generative-models), |
| [ControlNet](https://github.com/lllyasviel/ControlNet). |
| We acknowledge these open-source codes and models, and the website [CGTrader](https://www.cgtrader.com) for supporting 3D model free downloads. |
| We appreciate as well the valuable insights from the researchers |
| at the Shenzhen Institute of Advanced Technology and the |
| Shanghai AI Laboratory. |
|
|
|
|
| ## 📝 Citation |
| ```bib |
| @article{ding2024muses, |
| title={MUSES: 3D-Controllable Image Generation via Multi-Modal Agent Collaboration}, |
| author={Yanbo Ding and Shaobin Zhuang and Kunchang Li and Zhengrong Yue and Yu Qiao and Yali Wang}, |
| journal={arXiv preprint arXiv:2408.10605}, |
| year={2024}, |
| } |
| ``` |
|
|
