Skip to content

Latest commit

 

History

History
57 lines (49 loc) · 2.86 KB

README.md

File metadata and controls

57 lines (49 loc) · 2.86 KB

mv-diffusion

  • The purpose of the current directory is predicting source views.

Installation

Manually Install Using pip.

conda create --name geodream_mv python=3.8
conda activate geodream_mv
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
pip install inplace_abn
sudo apt-get install libsparsehash-dev
pip install git+https://github.com/mit-han-lab/[email protected]
pip install -r requirements.txt

Model Card

Manually download.

Model Weight Link Path
MVdream sd-v2.1-base-4view.pt GeoDream/mv-diffusion/MVDream/weight/sd-v2.1-base-4view.pt
zero123plus zero123plus-v1.1 GeoDream/mv-diffusion/zero123plus/weight/zero123plus-v1.1
zero123plus zero123plus-pipeline GeoDream/mv-diffusion/zero123plus/weight/zero123plus-pipeline
CLIP-ViT-H-14-laion2B-s32B-b79K CLIP-ViT-H-14-laion2B-s32B-b79K GeoDream/mv-diffusion/MVDream/CLIP-ViT-H-14-laion2B-s32B-b79K
Zero123-xl Zero123-xl GeoDream/mv-diffusion/One-2-3-45-by-view/zero123-xl.ckpt
Stable-zero123 Stable-zero123 GeoDream/mv-diffusion/One-2-3-45-by-view/stable_zero123.ckpt
SAM SAM GeoDream/mv-diffusion/One-2-3-45-by-view/sam_vit_h_4b8939.pth

Run

Construct cost volume by prompt

conda activate geodream_mv
cd GeoDream/mv-diffusion
sh step1-run-mv.sh "An astronaut riding a horse"
conda deactivate

. venv/bin/activate
sh step2-run-volume.sh "An astronaut riding a horse"

Construct cost volume by reference view

conda activate geodream_mv
cd GeoDream/mv-diffusion
# use zero123 
# Note : you should download `Zero123-xl` and `SAM`
sh run-volume-by-zero123.sh "An astronaut riding a horse" "ref_imges/demo.png"
# use stabale-zero123
# Note : you should download `Stable-zero123` and `SAM`
sh run-volume-by-sd-zero123.sh "An astronaut riding a horse" "ref_imges/demo.png"
conda deactivate

Acknowledgement

This repository is heavily based on Stable Diffusion, MVdream, One-2-3-45, zero123plus. We would like to thank the authors of these work for publicly releasing their code.