Project Page • Arxiv • Demo • FAQ • Citation
paint3d_vid_720p.mp4
Paint3D is a novel coarse-to-fine generative framework that is capable of producing high-resolution, lighting-less, and diverse 2K UV texture maps for untextured 3D meshes conditioned on text or image inputs.
Technical details
We present Paint3D, a novel coarse-to-fine generative framework that is capable of producing high-resolution, lighting-less, and diverse 2K UV texture maps for untextured 3D meshes conditioned on text or image inputs. The key challenge addressed is generating high-quality textures without embedded illumination information, which allows the textures to be re-lighted or re-edited within modern graphics pipelines. To achieve this, our method first leverages a pre-trained depth-aware 2D diffusion model to generate view-conditional images and perform multi-view texture fusion, producing an initial coarse texture map. However, as 2D models cannot fully represent 3D shapes and disable lighting effects, the coarse texture map exhibits incomplete areas and illumination artifacts. To resolve this, we train separate UV Inpainting and UVHD diffusion models specialized for the shape-aware refinement of incomplete areas and the removal of illumination artifacts. Through this coarse-to-fine process, Paint3D can produce high-quality 2K UV textures that maintain semantic consistency while being lighting-less, significantly advancing the state-of-the-art in texturing 3D objects.
- [2024/04/26] Upload code 🔥🔥🔥
- [2023/12/21] Upload paper and init project 🔥🔥🔥
The code is tested on Centos 7 with PyTorch 1.12.1 CUDA 11.6 installed. Please follow the following steps to setup environment.
# install python environment
conda env create -f environment.yaml
# install kaolin
pip install kaolin==0.13.0 -f https://nvidia-kaolin.s3.us-east-2.amazonaws.com/{TORCH_VER}_{CUDA_VER}.html
For UV-position controlnet, you can find it here.
To use the other ControlNet models, please download it from the hugging face page, and modify the controlnet path in the config file.
Then, you can generate coarse texture via:
python pipeline_paint3d_stage1.py \
--sd_config controlnet/config/depth_based_inpaint_template.yaml \
--render_config paint3d/config/train_config_paint3d.py \
--mesh_path demo/objs/Suzanne_monkey/Suzanne_monkey.obj \
--outdir outputs/stage1
and the refined texture via:
python pipeline_paint3d_stage2.py \
--sd_config controlnet/config/UV_based_inpaint_template.yaml \
--render_config paint3d/config/train_config_paint3d.py \
--mesh_path demo/objs/Suzanne_monkey/Suzanne_monkey.obj \
--texture_path outputs/stage1/res-0/albedo.png \
--outdir outputs/stage2
Optionally, you can also generate texture results with UV position controlnet only, for example:
python pipeline_UV_only.py \
--sd_config controlnet/config/UV_gen_template.yaml \
--render_config paint3d/config/train_config_paint3d.py \
--mesh_path demo/objs/teapot/scene.obj \
--outdir outputs/test_teapot
With a image condition, you can generate coarse texture via:
python pipeline_paint3d_stage1.py \
--sd_config controlnet/config/depth_based_inpaint_template.yaml \
--render_config paint3d/config/train_config_paint3d.py \
--mesh_path demo/objs/Suzanne_monkey/Suzanne_monkey.obj \
--prompt " " \
--ip_adapter_image_path demo/objs/Suzanne_monkey/img_prompt.png \
--outdir outputs/img_stage1
and the refined texture via:
python pipeline_paint3d_stage2.py \
--sd_config controlnet/config/UV_based_inpaint_template.yaml \
--render_config paint3d/config/train_config_paint3d.py \
--mesh_path demo/objs/Suzanne_monkey/Suzanne_monkey.obj \
--texture_path outputs/img_stage1/res-0/albedo.png \
--prompt " " \
--ip_adapter_image_path demo/objs/Suzanne_monkey/img_prompt.png \
--outdir outputs/img_stage2
For checkpoints in Civitai with only a .safetensor file, you can use the following script to convert and use them.
python tools/convert_original_stable_diffusion_to_diffusers.py \
--checkpoint_path YOUR_LOCAL.safetensors \
--dump_path model_cvt/ \
--from_safetensors
@misc{zeng2023paint3d,
title={Paint3D: Paint Anything 3D with Lighting-Less Texture Diffusion Models},
author={Xianfang Zeng and Xin Chen and Zhongqi Qi and Wen Liu and Zibo Zhao and Zhibin Wang and BIN FU and Yong Liu and Gang Yu},
year={2023},
eprint={2312.13913},
archivePrefix={arXiv},
primaryClass={cs.CV}
}Thanks to TEXTure, Text2Tex, Stable Diffusion and ControlNet, our code is partially borrowing from them. Our approach is inspired by MotionGPT, Michelangelo and DreamFusion.
This code is distributed under an Apache 2.0 LICENSE.
Note that our code depends on other libraries, including PyTorch3D and PyTorch Lightning, and uses datasets which each have their own respective licenses that must also be followed.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent