About the chamfer loss result

[yqf2000119]

Hi, thank you for your contribution. When I run your code, the final chamfer loss value is 0.0x, not 0.78 as shown in the paper. I used the same amount of experimental data as in the paper for the experiment, and the data came from Tallinn. I want to ask how the final loss function result of the paper completion part was obtained. I put the specific results in the screenshot.

[Timeless-H]

Three things: From my paper you will find that: (1) I did not use Tallinn data so even if you use my code, the results will definitely be different. (2) The 0.78 chamfer distant should be multiplied by 10^-3 as indicated in the table caption. Thus, the actual value is 0.00078. (3) Tallinn is the capital city, and am assuming has very complex/tall buildings which might degrade completion accuracy a little. Unless of course you have the compute resources to experiment with higher input sizes, output sizes, and/or scale (k) values.

I hope you find this helpful, best of luck.

Get Outlook for Androidhttps://aka.ms/AAb9ysg

---

From: yqf2000119 @.> Sent: Tuesday, September 24, 2024 6:50:48 AM To: geospatial-lab/APC2Mesh @.> Cc: Subscribed @.***> Subject: [geospatial-lab/APC2Mesh] About the chamfer loss result (Issue #3)

Hi, thank you for your contribution. When I run your code, the final chamfer loss value is 0.0x, not 0.78 as shown in the paper. I used the same amount of experimental data as in the paper for the experiment, and the data came from Tallinn. I want to ask how the final loss function result of the paper completion part was obtained. I put the specific results in the screenshot. Screenshot.from.2024-09-24.20-45-57.png (view on web)https://github.com/user-attachments/assets/7abd58a1-3731-48fd-9336-0be9f287c69a

— Reply to this email directly, view it on GitHubhttps://github.com/geospatial-lab/APC2Mesh/issues/3, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ASQZUXIUJOBTX3N6Y57V253ZYFN2NAVCNFSM6AAAAABOYHO26GVHI2DSMVQWIX3LMV43ASLTON2WKOZSGU2DKMZRHE2TGOI. You are receiving this because you are subscribed to this thread.Message ID: @.***>

[yqf2000119]

Three things: From my paper you will find that: (1) I did not use Tallinn data so even if you use my code, the results will definitely be different. (2) The 0.78 chamfer distant should be multiplied by 10^-3 as indicated in the table caption. Thus, the actual value is 0.00078. (3) Tallinn is the capital city, and am assuming has very complex/tall buildings which might degrade completion accuracy a little. Unless of course you have the compute resources to experiment with higher input sizes, output sizes, and/or scale (k) values. I hope you find this helpful, best of luck. Get Outlook for Androidhttps://aka.ms/AAb9ysg … ____ From: yqf2000119 @.> Sent: Tuesday, September 24, 2024 6:50:48 AM To: geospatial-lab/APC2Mesh @.> Cc: Subscribed @.> Subject: [geospatial-lab/APC2Mesh] About the chamfer loss result (Issue #3) Hi, thank you for your contribution. When I run your code, the final chamfer loss value is 0.0x, not 0.78 as shown in the paper. I used the same amount of experimental data as in the paper for the experiment, and the data came from Tallinn. I want to ask how the final loss function result of the paper completion part was obtained. I put the specific results in the screenshot. Screenshot.from.2024-09-24.20-45-57.png (view on web)https://github.com/user-attachments/assets/7abd58a1-3731-48fd-9336-0be9f287c69a — Reply to this email directly, view it on GitHub<#3>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ASQZUXIUJOBTX3N6Y57V253ZYFN2NAVCNFSM6AAAAABOYHO26GVHI2DSMVQWIX3LMV43ASLTON2WKOZSGU2DKMZRHE2TGOI. You are receiving this because you are subscribed to this thread.Message ID: @.>

thanks for your reply promptly, now I know my error. I misunderstood the chamfer distance result I thought it should multiplied by 10^-3 after 0.78 like 0.780, I also hope to use other data but the Building3D website only provides the Tallinn, In addition, my goal is to use the Tallinn as the pretrain model on APC2Mesh to reconstruction other cities I segment. I will try to think further about how to deal with the complex structure of Tallinn to achieve my goal. thanks for your answer my question again.

[Timeless-H]

@yqf2000119 I believe I provided a link to some more dataset other than Tallinn in the ReadMe.md file

[yqf2000119]

@Timeless-HThank you for the data download link you provided in the readme. I didn't read it carefully enough before. Thank you again for your prompt reply.