论文链接:https://arxiv.org/pdf/2310.15008.pdf

【AIGC大模型】跑通wonder3D (windows)-LMLPHP

windows10系统

显卡:NVIDIA rtx 2060 

一、安装anaconda 

二、安装CUDA  11.7 (CUDA Toolkit 11.7 Downloads | NVIDIA Developer) 和 cudnn 8.9.7(cuDNN Archive | NVIDIA Developer)库  CUDA选择自定义安装,其他环境变量配置啥的 可以参考下安装CUDA以及CUDNN(windows版)_windows安装cuda和cudnn-CSDN博客

【AIGC大模型】跑通wonder3D (windows)-LMLPHP

【AIGC大模型】跑通wonder3D (windows)-LMLPHP

三、(1)进入anaconda prompt 创建虚拟环境:conda create -n wonder3d  python=3.9  

conda create -n wonder3d  python=3.9  

(2)进入wonder3d 环境: conda activate wonder3d 

 conda activate wonder3d 

【AIGC大模型】跑通wonder3D (windows)-LMLPHP

【AIGC大模型】跑通wonder3D (windows)-LMLPHP

四、安装 pytorch 系列库 :

因为直接安装 到numpy的时候会报超时错误,所以先安装numpy 再安装torch 

 C:\ProgramData\Anaconda3\envs\wonder3d\python.exe -m pip install --upgrade pip -i http://pypi.douban.com/simple --trusted-host pypi.douban.com numpy


pip --default-timeout=6000 install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117  --trusted-host mirrors.aliyun.com

五、下载wonder3D 源码 并安装依赖项

安装一下 git 

conda install git

下载windows分支  

git clone https://github.com/xxlong0/Wonder3D.git -b main-windows

进入wonder3D目录

cd Wonder3D

【AIGC大模型】跑通wonder3D (windows)-LMLPHP

安装依赖项:

pip install -r requirements.txt -i http://pypi.douban.com/simple --trusted-host pypi.douban.com numpy

pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

【AIGC大模型】跑通wonder3D (windows)-LMLPHP

六、(1)下载Wonder3D的预训练模型 checkpoint . 把它放在 ckpts文件夹下.

Wonder3D
|-- ckpts
    |-- unet
    |-- scheduler.bin
    ...

(2) 修改./configs/mvdiffusion-joint-ortho-6views.yaml, 为 pretrained_model_name_or_path='./ckpts'

【AIGC大模型】跑通wonder3D (windows)-LMLPHP

(3)下载SAM大模型 SAM ,并把它放在sam_pt文件夹下

Wonder3D
|-- sam_pt
    |-- sam_vit_h_4b8939.pth

七、测试

python gradio_app.py
02-26 11:12