verlab
Clarification on MegaDepth Dataset Preparation & Broken Link
Namaste,
First off, thank you for creating and sharing this excellent feature matcher!
I have a question regarding the preparation of the MegaDepth dataset mentioned in your documentation. The setup guide refers to the LoFTR training instructions, which state:
"We use depth maps provided in the original MegaDepth dataset as well as undistorted images, corresponding camera intrinsics and extrinsics preprocessed by D2-Net..."
Based on this, I believe the correct procedure is:
Download MegaDepth_v1.tar.gz from the original MegaDepth project.
From this file, use only the depth maps.
Download the preprocessed images, camera intrinsics, and extrinsics from the D2-Net link and use them instead of the original images.
Could you please confirm if this understanding of the data preparation steps is correct?
Broken Data Link
I also wanted to report that the Google Drive link for the D2-Net preprocessed images appears to be broken (it returns a 404 error).
If you have an updated or alternative link to download these preprocessed files, could you please provide it?
Thank you for your time and help!
合十礼，
首先，非常感谢您创建并分享了这个优秀的特征匹配器！
我有一个关于文档中提到的 MegaDepth 数据集准备工作的问题。设置指南引用了LoFTR 训练说明，其中提到：
“我们使用原始 MegaDepth 数据集中提供的深度图，以及经D2-Net预处理后的无畸变图像、相应的相机内参和外参…… ”
基于以上分析，我认为正确的步骤是：
从 MegaDepth 原项目下载 MegaDepth_v1.tar.gz。
从该文件中，仅使用深度图。
从 D2-Net 链接下载预处理后的图像、相机内参和外参，并用它们代替原始图像。
请问您对数据准备步骤的理解是否正确？
数据链路故障
我还想报告一下，D2-Net 预处理图像的 Google Drive 链接似乎已损坏（返回 404 错误）。
如果您有更新的或备用的下载这些预处理文件的链接，能否请您提供一下？
感谢您抽出时间提供帮助！
Hello, did you solve it? I met this problem too
