search

SYSU-STAR / H2-Mapping

H2-Mapping: Real-time Dense Mapping Using Hierarchical Hybrid Representation (2023 RAL Best Paper Award)
GNU General Public License v3.0
362 stars 12 forks source link

Get bad mesh with our L515 data #16

Closed small-zeng closed 6 months ago

small-zeng commented 8 months ago

Thanks for your ncie work. After I collect data with L515 camera ，we use droid slam to estimate pose and reconstruction the scene with H2-mapping, why we get so bad mesh?

image

JIANG-CX commented 8 months ago

Several factors can influence the final results, such as the accuracy of the pose estimation and the duration of training. I apologize for not being able to directly identify the issue with your scene based on the provided mesh. However, I suggest comparing it to our provided dataset to identify any potential discrepancies or problems. Thank you for your understanding.

small-zeng commented 8 months ago

Several factors can influence the final results, such as the accuracy of the pose estimation and the duration of training. I apologize for not being able to directly identify the issue with your scene based on the provided mesh. However, I suggest comparing it to our provided dataset to identify any potential discrepancies or problems. Thank you for your understanding.

感谢你的神速回复，我看你的论文中提到了coarse tsdf， 我是否可以从代码中导出这个模型(不启动训练)，来验证我的位姿或者图片数据是否有问题

JIANG-CX commented 8 months ago

可以，你可以试一下把do_mapping函数都注释了，然后得到的mesh近似于coarse tsdf

获取 Outlook for iOShttps://aka.ms/o0ukef

发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 1:23:39 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16)

Several factors can influence the final results, such as the accuracy of the pose estimation and the duration of training. I apologize for not being able to directly identify the issue with your scene based on the provided mesh. However, I suggest comparing it to our provided dataset to identify any potential discrepancies or problems. Thank you for your understanding.

感谢你的神速回复，我看你的论文中提到了coarse tsdf， 我是否可以从代码中导出这个模型(不启动训练)，来验证我的位姿或者图片数据是否有问题

― Reply to this email directly, view it on GitHubhttps://github.com/SYSU-STAR/H2-Mapping/issues/16#issuecomment-1811828532, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNWLMUUTUH2UPYNHNZDYERGVXAVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHAZDQNJTGI. You are receiving this because you commented.Message ID: @.***>

small-zeng commented 8 months ago

可以，你可以试一下把do_mapping函数都注释了，然后得到的mesh近似于coarse tsdf 获取 Outlook for iOShttps://aka.ms/o0ukef … ____ 发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 1:23:39 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16) Several factors can influence the final results, such as the accuracy of the pose estimation and the duration of training. I apologize for not being able to directly identify the issue with your scene based on the provided mesh. However, I suggest comparing it to our provided dataset to identify any potential discrepancies or problems. Thank you for your understanding. 感谢你的神速回复，我看你的论文中提到了coarse tsdf， 我是否可以从代码中导出这个模型(不启动训练)，来验证我的位姿或者图片数据是否有问题 ― Reply to this email directly, view it on GitHub<#16 (comment)>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNWLMUUTUH2UPYNHNZDYERGVXAVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHAZDQNJTGI. You are receiving this because you commented.Message ID: @.***>

谢谢解答，我已经注释了do mapping， 得到以下mesh， 似乎很奇怪，噪点很多 image

JIANG-CX commented 8 months ago

请问你的max_depth设置的多少呢？可以试一下设置小一些，因为515过远的深度噪声较大。也可以看一下realsense viewer，是否合理设置了515的参数

获取 Outlook for iOShttps://aka.ms/o0ukef

发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 1:36:39 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16)

可以，你可以试一下把do_mapping函数都注释了，然后得到的mesh近似于coarse tsdf 获取 Outlook for iOShttps://aka.ms/o0ukef … ____ 发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 1:23:39 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16https://github.com/SYSU-STAR/H2-Mapping/issues/16) Several factors can influence the final results, such as the accuracy of the pose estimation and the duration of training. I apologize for not being able to directly identify the issue with your scene based on the provided mesh. However, I suggest comparing it to our provided dataset to identify any potential discrepancies or problems. Thank you for your understanding. 感谢你的神速回复，我看你的论文中提到了coarse tsdf， 我是否可以从代码中导出这个模型(不启动训练)，来验证我的位姿或者图片数据是否有问题 D Reply to this email directly, view it on GitHub<#16 (comment)https://github.com/SYSU-STAR/H2-Mapping/issues/16#issuecomment-1811828532>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNWLMUUTUH2UPYNHNZDYERGVXAVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHAZDQNJTGI. You are receiving this because you commented.Message ID: @.***>

谢谢解答，我已经注释了do mapping， 得到以下mesh， 似乎很奇怪，噪点很多 [image]https://user-images.githubusercontent.com/63596020/283014728-6a477c74-9549-4da7-914b-48f71e6c5f1d.png

― Reply to this email directly, view it on GitHubhttps://github.com/SYSU-STAR/H2-Mapping/issues/16#issuecomment-1811839093, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNXHEWINTO6G7GP4GW3YERIGPAVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHAZTSMBZGM. You are receiving this because you commented.Message ID: @.***>

small-zeng commented 8 months ago

max_depth 这个我设置的是3m， 然后515我使用的默认参数 max range 模式， 当我在您的配置中尝试将vox size从0.1m 提高到0.02m， 似乎噪声减少了，这是为什么呢？ image

image

image

JIANG-CX commented 8 months ago

请问前面那个图中的噪声是在天花板吗？感觉像是天花板处因为视角过大，点云比较稀疏，且有深度噪声导致的。我在使用稀疏深度测量，比如雷达的时候，也会出现类似的情况。因为我设置里voxel内的最少点云数量，voxel size变小，也许就过滤掉了

获取 Outlook for iOShttps://aka.ms/o0ukef

发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 1:54:38 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16)

max_depth 这个我设置的是3m， 然后515我使用的默认参数 max range 模式， 当我在您的配置中尝试将vox size从0.1m 提高到0.02m， 似乎噪声减少了，这是为什么呢？ [image]https://user-images.githubusercontent.com/63596020/283017604-015054ca-2e43-4b2a-abe7-1f4ed5f10f44.png

[image]https://user-images.githubusercontent.com/63596020/283017578-835a6b61-7849-48f4-923f-334e23b134d7.png

[image]https://user-images.githubusercontent.com/63596020/283017653-eacf8dee-72f2-45d1-86bd-31689f6a4d01.png

― Reply to this email directly, view it on GitHubhttps://github.com/SYSU-STAR/H2-Mapping/issues/16#issuecomment-1811852093, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNSQKHYCKULTXLNYEHLYERKJ5AVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHA2TEMBZGM. You are receiving this because you commented.Message ID: @.***>

small-zeng commented 8 months ago

请问前面那个图中的噪声是在天花板吗？感觉像是天花板处因为视角过大，点云比较稀疏，且有深度噪声导致的。我在使用稀疏深度测量，比如雷达的时候，也会出现类似的情况。因为我设置里voxel内的最少点云数量，voxel size变小，也许就过滤掉了 获取 Outlook for iOShttps://aka.ms/o0ukef … ____ 发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 1:54:38 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16) max_depth 这个我设置的是3m， 然后515我使用的默认参数 max range 模式， 当我在您的配置中尝试将vox size从0.1m 提高到0.02m， 似乎噪声减少了，这是为什么呢？ [image]https://user-images.githubusercontent.com/63596020/283017604-015054ca-2e43-4b2a-abe7-1f4ed5f10f44.png [image]https://user-images.githubusercontent.com/63596020/283017578-835a6b61-7849-48f4-923f-334e23b134d7.png [image]https://user-images.githubusercontent.com/63596020/283017653-eacf8dee-72f2-45d1-86bd-31689f6a4d01.png ― Reply to this email directly, view it on GitHub<#16 (comment)>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNSQKHYCKULTXLNYEHLYERKJ5AVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHA2TEMBZGM. You are receiving this because you commented.Message ID: @.***>

前面那个图是看向墙和地面，应该是因为视角遮挡问题存在一些稀疏点

JIANG-CX commented 8 months ago

有可能🤔

获取 Outlook for iOShttps://aka.ms/o0ukef

发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 2:11:44 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16)

请问前面那个图中的噪声是在天花板吗？感觉像是天花板处因为视角过大，点云比较稀疏，且有深度噪声导致的。我在使用稀疏深度测量，比如雷达的时候，也会出现类似的情况。因为我设置里voxel内的最少点云数量，voxel size变小，也许就过滤掉了 获取 Outlook for iOShttps://aka.ms/o0ukef … ____ 发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 1:54:38 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16https://github.com/SYSU-STAR/H2-Mapping/issues/16) max_depth 这个我设置的是3m， 然后515我使用的默认参数 max range 模式， 当我在您的配置中尝试将vox size从0.1m 提高到0.02m， 似乎噪声减少了，这是为什么呢？ [image]https://user-images.githubusercontent.com/63596020/283017604-015054ca-2e43-4b2a-abe7-1f4ed5f10f44.png [image]https://user-images.githubusercontent.com/63596020/283017578-835a6b61-7849-48f4-923f-334e23b134d7.png [image]https://user-images.githubusercontent.com/63596020/283017653-eacf8dee-72f2-45d1-86bd-31689f6a4d01.png ― Reply to this email directly, view it on GitHub<#16 (comment)https://github.com/SYSU-STAR/H2-Mapping/issues/16#issuecomment-1811852093>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNSQKHYCKULTXLNYEHLYERKJ5AVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHA2TEMBZGM. You are receiving this because you commented.Message ID: @.***>

前面那个图是看向墙和地面，应该是因为视角遮挡问题存在一些稀疏点

— Reply to this email directly, view it on GitHubhttps://github.com/SYSU-STAR/H2-Mapping/issues/16#issuecomment-1811865473, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNRSNNRH5VUAYMDBLK3YERMKBAVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHA3DKNBXGM. You are receiving this because you commented.Message ID: @.***>

small-zeng commented 8 months ago

那还有什么参数可以调整让他变好吗

small-zeng commented 8 months ago

我感觉这个噪点如果一直存在，后面训练一直不会好。请问您的tsdf 融合的结果是不是没有这么多噪点

JIANG-CX commented 8 months ago

可以把515对噪声的滤波再开大一点。然后现在这个方案对pose还是比较敏感的，看你第一个mesh，似乎沙发那边呈现了一些pose的问题。可以对比一下提供的bag，看下噪声水平。

获取 Outlook for iOShttps://aka.ms/o0ukef

发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 2:14:34 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16)

那还有什么参数可以调整让他变好吗

― Reply to this email directly, view it on GitHubhttps://github.com/SYSU-STAR/H2-Mapping/issues/16#issuecomment-1811867800, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNX566MZZB4PUR4DNEDYERMUVAVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHA3DOOBQGA. You are receiving this because you commented.Message ID: @.***>

small-zeng commented 8 months ago

好的感谢，请问这个滤波参数对应配置文件的哪个变量呢


---- 回复的原邮件 ----
发件人 JIANG ***@***.***>
日期 2023年11月15日 14:19
收件人 ***@***.***>
抄送至 Jing ***@***.***>、***@***.***>
主题 Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data  (Issue #16)

可以把515对噪声的滤波再开大一点。然后现在这个方案对pose还是比较敏感的，看你第一个mesh，似乎沙发那边呈现了一些pose的问题。可以对比一下提供的bag，看下噪声水平。

获取 Outlook for iOS<https://aka.ms/o0ukef>
________________________________
发件人: Jing Zeng ***@***.***>
发送时间: Wednesday, November 15, 2023 2:14:34 PM
收件人: SYSU-STAR/H2-Mapping ***@***.***>
抄送: JIANG Chenxing ***@***.***>; Comment ***@***.***>
主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16)


那还有什么参数可以调整让他变好吗

―
Reply to this email directly, view it on GitHub<https://github.com/SYSU-STAR/H2-Mapping/issues/16#issuecomment-1811867800>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/APMKKNX566MZZB4PUR4DNEDYERMUVAVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHA3DOOBQGA>.
You are receiving this because you commented.Message ID: ***@***.***>

—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: <SYSU-STAR/H2-Mapping/issues/16/1811872239@github.com>

JIANG-CX commented 8 months ago

Noise Filtering。https://github.com/SYSU-STAR/H2-Mapping/issues/8#issuecomment-1760985423这里是我设置的config文件，可以参考。

small-zeng commented 8 months ago

明白了，这里的滤波是指在修改L515的launch 文件把

JIANG-CX commented 8 months ago

嗯，可以尝试一下

获取 Outlook for iOShttps://aka.ms/o0ukef

发件人: Jing Zeng @.> 发送时间: Wednesday, November 15, 2023 2:44:09 PM 收件人: SYSU-STAR/H2-Mapping @.> 抄送: JIANG Chenxing @.>; Comment @.> 主题: Re: [SYSU-STAR/H2-Mapping] Get bad mesh with our L515 data (Issue #16)

明白了，这里的滤波是指在修改L515的launch 文件把

― Reply to this email directly, view it on GitHubhttps://github.com/SYSU-STAR/H2-Mapping/issues/16#issuecomment-1811891432, or unsubscribehttps://github.com/notifications/unsubscribe-auth/APMKKNQZSDAT4ML577OIAVDYERQDTAVCNFSM6AAAAAA7KTTLPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMJRHA4TCNBTGI. You are receiving this because you commented.Message ID: @.***>