search

rebecca0011 / UAVStereo

23 stars 0 forks source link

Question about pose data. #3

Open euncheolChoi opened 1 year ago

euncheolChoi commented 1 year ago

Thank you for sharing great dataset. I haven't downloaded it yet as it's hard to download through the Baidu link. Because i'd like to use this dataset for Localization task, I want to ask whether does the dataset you provided also include Pose data ( [x, y, z, roll, pitch, yaw] )?

rebecca0011 commented 1 year ago

Thank you for sharing great dataset. I haven't downloaded it yet as it's hard to download through the Baidu link. Because i'd like to use this dataset for Localization task, I want to ask whether does the dataset you provided also include Pose data ( [x, y, z, roll, pitch, yaw] )?

The dataset does not contain position and orientation information. The synthetic data is generated using local coordinates, and the position and orientation of the real data is removed before publication

euncheolChoi commented 1 year ago

Thanks you for reply. I am interested in Real data. If you still have Pose information, I wonder if you can share it.

rebecca0011 commented 1 year ago

Thanks you for reply. I am interested in Real data. If you still have Pose information, I wonder if you can share it.

Sorry, we don't intend to disclose real data position and orientation information.

euncheolChoi commented 1 year ago

All right. Anyway, thanks for the great dataset.

luoying19940423 commented 1 year ago

你好，我看到Forest测试集中很多数据似乎是重复的，数据数量似乎与论文中并不一样？请问能否方便解惑

rebecca0011 commented 1 year ago

你好，我看到Forest测试集中很多数据似乎是重复的，数据数量似乎与论文中并不一样？请问能否方便解惑

文件夹中包含的数据不是重复的，影像看起来重复可能是因为：1.文件夹中包含同一地区不同高度的影像；2.forest重建数据包含很多重复纹理，看起来差异不大。

因为数据包含特殊区域，所以我们在上传之前删除掉了相关区域的影像，引起了数量的不一致。

luoying19940423 commented 1 year ago

谢谢回复！

luoying19940423 commented 1 year ago

还想请问您低分辨率训练时PSM-NET的匹配代价体最大搜索视差是保持原值还是做了更改，请问能否方便解惑？

rebecca0011 commented 1 year ago

还想请问您低分辨率训练时PSM-NET的匹配代价体最大搜索视差是保持原值还是做了更改，请问能否方便解惑？

最大视差搜索视差的设置与数据视差分布相关，UAVStereo中数据中视差大于KITTI和Sceneflow，实验时的设置为384px.

luoying19940423 commented 1 year ago

谢谢回答！

luoying19940423 commented 11 months ago

还想请问您低分辨率训练时PSM-NET的匹配代价体最大搜索视差是保持原值还是做了更改，请问能否方便解惑？

最大视差搜索视差的设置与数据视差分布相关，UAVStereo中数据中视差大于KITTI和Sceneflow，实验时的设置为384px.

我想请问训练好的网络模型在低分辨率测试集上预测时，视差大于384px的位置是否被屏蔽掉

rebecca0011 commented 11 months ago

还想请问您低分辨率训练时PSM-NET的匹配代价体最大搜索视差是保持原值还是做了更改，请问能否方便解惑？

最大视差搜索视差的设置与数据视差分布相关，UAVStereo中数据中视差大于KITTI和Sceneflow，实验时的设置为384px.

我想请问训练好的网络模型在低分辨率测试集上预测时，视差大于384px的位置是否被屏蔽掉

我们数据集在设计的时候没有屏蔽掉大于384px的视差。 一般在进行网络精度验证时，可以生成mask，去掉视差大于384px的像素。

luoying19940423 commented 11 months ago

还想请问您低分辨率训练时PSM-NET的匹配代价体最大搜索视差是保持原值还是做了更改，请问能否方便解惑？

最大视差搜索视差的设置与数据视差分布相关，UAVStereo中数据中视差大于KITTI和Sceneflow，实验时的设置为384px.

我想请问训练好的网络模型在低分辨率测试集上预测时，视差大于384px的位置是否被屏蔽掉

我们数据集在设计的时候没有屏蔽掉大于384px的视差。 一般在进行网络精度验证时，可以生成mask，去掉视差大于384px的像素。

非常感谢回答！