许可协议: apache-2.0
标签:

• 视觉
• 深度估计

展示组件:

• 示例图片: https://huggingface.co/datasets/mishig/sample_images/resolve/main/tiger.jpg
  示例标题: 老虎
• 示例图片: https://huggingface.co/datasets/mishig/sample_images/resolve/main/teapot.jpg
  示例标题: 茶壶
• 示例图片: https://huggingface.co/datasets/mishig/sample_images/resolve/main/palace.jpg
  示例标题: 宫殿

基于KITTI微调的GLPN模型

全局-局部路径网络（GLPN）在KITTI数据集上训练的单目深度估计模型。该模型由Kim等人在论文《采用垂直CutDepth技术的单目深度估计全局-局部路径网络》中提出，并首次发布于此代码库。

免责声明：GLPN发布团队未提供该模型的说明卡片，本卡片由Hugging Face团队撰写。

模型描述

GLPN采用SegFormer作为主干网络，顶部添加轻量级头部进行深度估计。

应用场景与限制

该模型可直接用于单目深度估计任务。访问模型中心可查看针对特定任务微调的版本。

使用方法

使用方式如下：

from transformers import GLPNImageProcessor, GLPNForDepthEstimation
import torch
import numpy as np
from PIL import Image
import requests

url = "http://images.cocodataset.org/val2017/000000039769.jpg"
image = Image.open(requests.get(url, stream=True).raw)

processor = GLPNImageProcessor.from_pretrained("vinvino02/glpn-kitti")
model = GLPNForDepthEstimation.from_pretrained("vinvino02/glpn-kitti")

# 为模型准备图像
inputs = processor(images=image, return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)
    predicted_depth = outputs.predicted_depth

# 插值至原始尺寸
prediction = torch.nn.functional.interpolate(
    predicted_depth.unsqueeze(1),
    size=image.size[::-1],
    mode="bicubic",
    align_corners=False,
)

# 可视化预测结果
output = prediction.squeeze().cpu().numpy()
formatted = (output * 255 / np.max(output)).astype("uint8")
depth = Image.fromarray(formatted)

更多代码示例详见文档。

引用信息

@article{DBLP:journals/corr/abs-2201-07436,
  author    = {Doyeon Kim and
               Woonghyun Ga and
               Pyunghwan Ahn and
               Donggyu Joo and
               Sehwan Chun and
               Junmo Kim},
  title     = {Global-Local Path Networks for Monocular Depth Estimation with Vertical
               CutDepth},
  journal   = {CoRR},
  volume    = {abs/2201.07436},
  year      = {2022},
  url       = {https://arxiv.org/abs/2201.07436},
  eprinttype = {arXiv},
  eprint    = {2201.07436},
  timestamp = {Fri, 21 Jan 2022 13:57:15 +0100},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2201-07436.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}