图生3d 图生全景 学习笔记

目录

instantsplat

Aluciddreamer

ZoeDepth

会自动下载模型：

图生全景图SD-T2I-360PanoImage：

---

---

instantsplat

Sparse-view SfM-free Gaussian Splatting in Seconds

稀疏视图无SfM高斯喷洒

GitHub - NVlabs/InstantSplat: InstantSplat: Sparse-view SfM-free Gaussian Splatting in Seconds

Aluciddreamer

GitHub - luciddreamer-cvlab/LucidDreamer: Official code for the paper "LucidDreamer: Domain-free Generation of 3D Gaussian Splatting Scenes".

解析器添加参数（‘--campath_gen’，‘-cg’），类型为字符串，默认值为‘rotate360’，可选值为 [‘lookdown’，‘lookaround’，‘rotate360’]，帮助信息为 “用于场景生成的相机外参轨迹”。

解析器添加参数（‘--campath_render’，‘-cr’），类型为字符串，默认值为‘back_and_forth’，可选值为 [‘back_and_forth’，‘llff’，‘headbanging’]，帮助信息为 “用于视频渲染的相机外参轨迹”。

ZoeDepth

引用地址：

GitHub - isl-org/ZoeDepth: Metric depth estimation from a single image

演示地址：

https://huggingface.co/spaces/shariqfarooq/ZoeDepth

模型下载地址：

Releases · isl-org/ZoeDepth · GitHub

会自动下载模型：

self.d_model = torch.hub.load('./ZoeDepth', 'ZoeD_N', source='local', pretrained=True).to('cuda')

下载路径：

/mnt/pfs/models/torch/hub/intel-isl_MiDaS_master    Using cache found in

/mnt/pfs/models/torch/hub/checkpoints

图生全景图SD-T2I-360PanoImage：

pip install numpy==1.23.2

import sys
import os
os.chdir(os.path.dirname(os.path.abspath(__file__)))import torchcurrent_dir = os.path.dirname(os.path.abspath(__file__))paths = [os.path.abspath(__file__).split('scripts')[0]]
print('current_dir',current_dir)
paths.append(os.path.abspath(os.path.join(current_dir, 'src')))for path in paths:sys.path.insert(0, path)os.environ['PYTHONPATH'] = (os.environ.get('PYTHONPATH', '') + ':' + path).strip(':')import torch
from diffusers.utils import load_image
from img2panoimg import Image2360PanoramaImagePipelineimage = load_image("./data/i2p-image.jpg").resize((512, 512))
mask = load_image("./data/i2p-mask.jpg")prompt = 'The office room'# for <16GB gpu
input = {'prompt': prompt, 'image': image, 'mask': mask, 'upscale': False}# for >16GB gpu (24GB at least)
# the similarity with the input image is poor because of the super-resolution steps. It should be improved.
# input = {'prompt': prompt, 'image': image, 'mask': mask, 'upscale': True}model_id = 'models'
img2panoimg = Image2360PanoramaImagePipeline(model_id, torch_dtype=torch.float16)
output = img2panoimg(input)
output.save('result.png')