Hunyuan3D-1 / scripts /image_to_3d_lite_separately.sh
Huiwenshi's picture
Upload folder using huggingface_hub
e3e5f9e verified
raw
history blame
1.49 kB
# !/usr/bin/bash
# ------------------------------------
# stage | Minimum Memory Support
# ------------------------------------
# text_to_image | ~14G
# ------------------------------------
# image_to_views | lite ~8G
# image_to_views | std ~16G
# ------------------------------------
# views_to_mesh | ~10G (90000 faces)
# ------------------------------------
# Usage:
# bash scripts/image_to_3d_lite_separately.sh ./demos/example_000.png ./outputs/test
# # need Memory > 10G
rgb_path=$1
save_folder=$2
# init
use_lite=true
do_texture_mapping=true
max_faces_num=90000
mkdir -p $save_folder
python infer/removebg.py \
--rgb_path $rgb_path \
--output_rgba_path $save_folder/img_nobg.png \
&& \
python infer/image_to_views.py \
--rgba_path $save_folder/img_nobg.png \
--output_views_path $save_folder/views.jpg \
--output_cond_path $save_folder/cond.jpg \
--seed 0 \
--steps 50 \
--device "cuda:0" \
--use_lite $use_lite \
&& \
python infer/views_to_mesh.py \
--views_path $save_folder/views.jpg \
--cond_path $save_folder/cond.jpg \
--save_folder $save_folder \
--max_faces_num $max_faces_num \
--mv23d_cfg_path ./svrm/configs/svrm.yaml \
--mv23d_ckt_path ./weights/svrm/svrm.safetensors \
--device "cuda:0" \
--use_lite $use_lite \
--do_texture_mapping $do_texture_mapping \
&& \
python infer/gif_render.py \
--mesh_path $save_folder/mesh.obj \
--output_gif_path $save_folder/output.gif