首页 > 其他分享 >datasets for stereo depth

datasets for stereo depth

时间:2024-05-31 17:34:19浏览次数:21  
标签:disp stereo datasets https header depth file data match

  1. CRE dateset
# 0,1,2,3
https://data.megengine.org.cn/research/crestereo/dataset/tree/0.tar
https://data.megengine.org.cn/research/crestereo/dataset/shapenet/0.tar
https://data.megengine.org.cn/research/crestereo/dataset/reflective/0.tar
https://data.megengine.org.cn/research/crestereo/dataset/hole/0.tar


def get_disp(disp_path):
    disp = cv2.imread(disp_path, cv2.IMREAD_UNCHANGED)
    return disp.astype(np.float32) / 32

	
  1. Falling Things
http://research.nvidia.com/publication/2018-06_Falling-Things
https://drive.google.com/file/d/1y4h9T6D9rf6dAmsRwEtfzJdcghCnI_01/view
[BT](magnet:?xt=urn:btih:5643313104D5000D183250EC341D6291FBC89554)

depth数据 
Depth along the optical axis (in 0.1 mm increments)

估算以及读txt验证 B = 600
f = 480 / tan(32°) = 768.16058349609375

disp = B*f/depth

  1. Sceneflow

https://lmb.informatik.uni-freiburg.de/data/SceneFlowDatasets_CVPR16/Release_april16/data/FlyingThings3D/raw_data/flyingthings3d__frames_cleanpass.tar
https://lmb.informatik.uni-freiburg.de/data/SceneFlowDatasets_CVPR16/Release_april16/data/FlyingThings3D/derived_data/flyingthings3d__disparity.tar.bz2
  1. Sintel
http://sintel.is.tue.mpg.de/downloads

disp = cv2.imread("disp..", cv2.IMREAD_UNCHANGED).astype(np.float64)
disp_float = disp[:, :, 2] * 4 + disp[:, :, 1] / (2 ** 6) + disp[:, :, 0] / (2 ** 14)

5.others
[InStereo2K] https://github.com/YuhuaXu/StereoDataset
disp = u16 / 100.

Read pfm files
import re
import sys
import numpy as np

def read_pfm(filename):
    with open(filename, 'rb') as file:
        # header
        header = file.readline().rstrip()
        if (sys.version[0]) == '3':
            header = header.decode('utf-8')
        if header == 'PF':
            color = True
        elif header == 'Pf':
            color = False
        else:
            raise Exception('Not a PFM file.')

        # width height
        if (sys.version[0]) == '3':
            dim_match = re.match(r'^(\d+)\s(\d+)\s$', file.readline().decode('utf-8'))
        else:
            dim_match = re.match(r'^(\d+)\s(\d+)\s$', file.readline())
        if dim_match:
            width, height = map(int, dim_match.groups())
        else:
            raise Exception('Malformed PFM header.')

        # scale
        if (sys.version[0]) == '3':
            scale = float(file.readline().rstrip().decode('utf-8'))
        else:
            scale = float(file.readline().rstrip())

        # endian
        if scale < 0:  # little-endian
            endian = '<'
            scale = -scale
        else:
            endian = '>'  # big-endian

        # data
        data = np.fromfile(file, endian + 'f')
    shape = (height, width, 3) if color else (height, width)

    data = np.reshape(data, shape)
    data = np.flipud(data)
    return data, scale

标签:disp,stereo,datasets,https,header,depth,file,data,match
From: https://www.cnblogs.com/wioponsen/p/18224961

相关文章

  • CompareBinaryTreeDepth
    /*******************************************************************************************************@filename: :CompareBinaryTreeDepth*@brief :采用递归的方式来计算二叉树的深度*@author :[email protected]*@date :2024/05/03*......
  • 深度优先搜索 Depth First Search (DFS)
    本篇篇幅较长,请做好心理准备!共三章节:1.深搜入门(一维方向数字选数类)2.深搜入门(二维方向迷宫类)3.深搜进阶(迷宫类问题--最少步数和输出路径)第一章:深搜入门(一维方向数字选数类)前置知识:函数、递归为了保证学习效果,请保证已经掌握前置知识之后,再来学习本章节!深度优......
  • An Analysis of Sequential Recommendation Datasets
    目录概统计角度论证实验论证代码WoolridgeD.,WilnerS.andGlickM.Ananalysisofsequentialrecommendationdatasets.PERSPECTIVES,2021.概本文讨论了MovieLens系列数据集是否适用于序列推荐.统计角度论证作者为了论证MovieLens不适合作为序列推荐数据集,......
  • Learning Imbalanced Datasets with Label-Distribution-Aware Margin Loss
    省去冗长的数学证明,直接看文章的贡献:提出了新的Loss函数以及延迟re-weighting的trick。并在多个数据集,包括情感分类、图像分类进行实验。Motivation&Methods:LDAM(Label-Distribution-AwareMargie)Losstailclasses的信息基本上较少,而且部署的模型通常很大,因此对tailclasse......
  • Stereotyped Writing(二)
    写出你用过的linux命令,举例说明其中的几个的具体用法。命令含义shutdown-hnow立刻关机shutdown-rnow立刻重启cd切换路径ls查看当前路径下的所有目录和文件mkdirtools在当前目录下创建一个名为tools的目录rm文件名删除当前目录下的文件......
  • Stereotyped Writing(一)
    线程的状态?新建(NEW):线程被创建但并未启动可运行(RUNNABLE):线程可以在任意时刻运行,也可能正在等待CPU分配时间片堵塞(BLOCKED):线程被阻止执行,因为它正在等待监视器锁定,其他线程正在占用所需的锁定,因此线程被堵塞等待(WAITING):线程进入等待状态,直到其他线程显示的唤醒他,线程可......
  • Characterizing Graph Datasets for Node Classification Homophily-Heterophily Dich
    目录概符号说明Popularhomophilymeasures理想的准则现有的metrics的分析PlatonovO.,KuznedelevD.,BabenkoA.andProkhorenkovaL.Characterizinggraphdatasetsfornodeclassification:homophily-heterophilydichotomyandbeyond.NIPS,2023.概阐述合理的......
  • OpenGL中的Frame Buffer、Depth Buffer、Color Buffer、Stencil Buffer
    FrameBufferFrameBuffer:是一块buffer(即内存),存储了一帧的buffer数据。从数据结构的角度来看,此处的FrameBuffer并不是一个真正意义上的buffer,其存了一些指针,分别指向DepthBuffer、ColorBuffer、StencilBuffer、Texture、RBO(RenderBufferObject)等。Wiki中的解释:https......
  • 2024AAAI_SGNet Structure Guided Network via Gradient-Frequency Awareness for Dep
    1.任务描述: 给定输入LR深度图和HRRGB图像,引导DSR目的是在ground-truth深度图监督的条件下,预测HR深度图2.Network本文提出的SGNet主要包括两部分,即梯度校准模块(GCM)和频率感知模块(FAM)。首先将RGB图像和上采样后的LR深度图送入到GCM,利用RGB丰富的梯度信息在梯度域中......
  • scikit-learn.datasets 机器学习库
    scikit-learn是一个用于Python的机器学习库,提供了大量用于数据挖掘和数据分析的工具。以下是对这些函数和方法的简要描述:clear_data_home:清除数据集目录的内容。dump_svmlight_file:将数据集保存为SVMLight格式的文件。fetch_20newsgroups:下载20个新闻组的文本数据集。f......