search

fvasluianu-cvl / SSSR

3 stars 0 forks source link

Table 5 MaskShadowGAN[16] is 7.32 in ISTD and the result in USR #1

Open nachifur opened 4 years ago

nachifur commented 4 years ago

Your work is great. ButI I have some questions about some content of the paper. image MaskShadowGAN[16] is 7.61 in ISTD (unpaired). Maybe 7.32 in Table5 is a small error. image And I am confused about the result in USR test images (blue box). Because I don't find the same result in MaskShadowGAN[16]. MaskShadowGAN[16]: As the unpaired USR dataset has no ground truths, we conducted a user study to evaluate the shadow removal results. How do you get the results of the evaluation metrics, like RMSE, PSNR? Looking forward to your reply.

fvasluianu-cvl commented 4 years ago

The results for MaskShadowGAN were produced as a local experiment, so maybe we were lucky with a better value.

The RMSE values ware computed for the reconstruction cost, compared to cycleGAN and MaskShadowGAN.

From: William Liu notifications@github.com Sent: Friday, October 30, 2020 4:01:59 AM To: fvasluianu-cvl/SSSR Cc: Subscribed Subject: [fvasluianu-cvl/SSSR] Table 5 MaskShadowGAN[16] is 7.32 in ISTD and the result in USR (#1)

Your work is great. ButI I have some questions about some content of the paper. [image]https://user-images.githubusercontent.com/23144348/97653868-bf33de80-1a9c-11eb-8c89-137a9b7dd2fb.png MaskShadowGAN[16] is 7.61 in ISTD (unpaired). Maybe 7.32 in Table5 is a small error. [image]https://user-images.githubusercontent.com/23144348/97653905-cf4bbe00-1a9c-11eb-9af6-530dc29048f5.png And I am confused about the result in USR test images (blue box). Because I don't find the same result in MaskShadowGAN[16]. MaskShadowGAN[16]: As the unpaired USR dataset has no ground truths, we conducted a user study to evaluate the shadow removal results. How do you get the results of the evaluation metrics, like RMSE, PSNR? Looking forward to your reply.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/fvasluianu-cvl/SSSR/issues/1, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AQCVKPMA5UJM5H7CMA4RY3DSNIUCPANCNFSM4TEPMEXQ.

nachifur commented 4 years ago

Sincerely thank you for your reply. When I first read, I thought you were quoting data from MaskShadowGAN[16]. I thought you quote the data wrong, after all, Table 3 in MaskShadowGAN[16] also appeared 7.32 in SRD dataset. If it’s the result of your own experiment, I am not confused now.

nachifur commented 4 years ago

Why are the data in Table 4 and Table 5 different? (under unpaired setting) Looking forward to your reply. image image

fvasluianu-cvl commented 4 years ago

For the ablative setting of the unpaired setup we limited the amount of training time until an early-convergence is reached.

I hope this answers to your question.

Thanks.

F. Vasluianu

From: William Liu notifications@github.com Sent: Friday, October 30, 2020 4:08:49 PM To: fvasluianu-cvl/SSSR Cc: Vasluianu Florin; Comment Subject: Re: [fvasluianu-cvl/SSSR] Table 5 MaskShadowGAN[16] is 7.32 in ISTD and the result in USR (#1)

Why are the data in Table 4 and Table 5 different? (under unpaired setting) Looking forward to your reply. [image]https://user-images.githubusercontent.com/23144348/97721546-87f91800-1b04-11eb-989b-998aaa59acbb.png [image]https://user-images.githubusercontent.com/23144348/97721579-947d7080-1b04-11eb-9493-5b89cddfc9ac.png

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/fvasluianu-cvl/SSSR/issues/1#issuecomment-719609037, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AQCVKPJP5WBS65XQBIMS56DSNLJIDANCNFSM4TEPMEXQ.

nachifur commented 4 years ago

Sincerely thank you for your reply. I am not confused now.-------- 原始邮件 --------发件人： fvasluianu-cvl notifications@github.com日期： 2020年10月30日周五 半夜11:52收件人： fvasluianu-cvl/SSSR SSSR@noreply.github.com抄送： William Liu 1476879092@qq.com, Author author@noreply.github.com主 题： Re: [fvasluianu-cvl/SSSR] Table 5 MaskShadowGAN[16] is 7.32 in ISTD and the result in USR (#1)

For the ablative setting of the unpaired setup we limited the amount of training time until an early-convergence is reached.

I hope this answers to your question.

Thanks.

F. Vasluianu

From: William Liu notifications@github.com

Sent: Friday, October 30, 2020 4:08:49 PM

To: fvasluianu-cvl/SSSR

Cc: Vasluianu Florin; Comment

Subject: Re: [fvasluianu-cvl/SSSR] Table 5 MaskShadowGAN[16] is 7.32 in ISTD and the result in USR (#1)

Why are the data in Table 4 and Table 5 different? (under unpaired setting) Looking forward to your reply.

[image]https://user-images.githubusercontent.com/23144348/97721546-87f91800-1b04-11eb-989b-998aaa59acbb.png

[image]https://user-images.githubusercontent.com/23144348/97721579-947d7080-1b04-11eb-9493-5b89cddfc9ac.png

—

You are receiving this because you commented.

Reply to this email directly, view it on GitHubhttps://github.com/fvasluianu-cvl/SSSR/issues/1#issuecomment-719609037, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AQCVKPJP5WBS65XQBIMS56DSNLJIDANCNFSM4TEPMEXQ.

—You are receiving this because you authored the thread.Reply to this email directly, view it on GitHub, or unsubscribe.

nachifur commented 4 years ago

Whether all data in Table 5 and Table 6 was computed for the reconstruction cost？Which result is the metric between the shadow image and the image after shadow removal. We care about these results related to the performance of shadow removal. Looking forward to your reply. image

The results for MaskShadowGAN were produced as a local experiment, so maybe we were lucky with a better value. The RMSE values ware computed for the reconstruction cost, compared to cycleGAN and MaskShadowGAN. … ____ From: William Liu notifications@github.com Sent: Friday, October 30, 2020 4:01:59 AM To: fvasluianu-cvl/SSSR Cc: Subscribed Subject: [fvasluianu-cvl/SSSR] Table 5 MaskShadowGAN[16] is 7.32 in ISTD and the result in USR (#1) Your work is great. ButI I have some questions about some content of the paper. [image]https://user-images.githubusercontent.com/23144348/97653868-bf33de80-1a9c-11eb-8c89-137a9b7dd2fb.png MaskShadowGAN[16] is 7.61 in ISTD (unpaired). Maybe 7.32 in Table5 is a small error. [image]https://user-images.githubusercontent.com/23144348/97653905-cf4bbe00-1a9c-11eb-9af6-530dc29048f5.png And I am confused about the result in USR test images (blue box). Because I don't find the same result in MaskShadowGAN[16]. MaskShadowGAN[16]: As the unpaired USR dataset has no ground truths, we conducted a user study to evaluate the shadow removal results. How do you get the results of the evaluation metrics, like RMSE, PSNR? Looking forward to your reply. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<#1>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AQCVKPMA5UJM5H7CMA4RY3DSNIUCPANCNFSM4TEPMEXQ.

fvasluianu-cvl commented 4 years ago

For the unpaired training, since it's impossible to have a metric assessing "shadowness", we reported the reconstruction loss compared to other methods.

From: William Liu notifications@github.com Sent: Monday, November 2, 2020 1:36:18 PM To: fvasluianu-cvl/SSSR Cc: Vasluianu Florin; Comment Subject: Re: [fvasluianu-cvl/SSSR] Table 5 MaskShadowGAN[16] is 7.32 in ISTD and the result in USR (#1)

Whether all data in Table 5 and Table 6 was computed for the reconstruction cost？Which result is the metric between the shadow image and the image after shadow removal. We care about these results related to the performance of shadow removal. Looking forward to your reply. [image]https://user-images.githubusercontent.com/23144348/97867764-6dfc4700-1d49-11eb-98dc-5dbb92558aca.png

The results for MaskShadowGAN were produced as a local experiment, so maybe we were lucky with a better value. The RMSE values ware computed for the reconstruction cost, compared to cycleGAN and MaskShadowGAN. … ____ From: William Liu notifications@github.commailto:notifications@github.com Sent: Friday, October 30, 2020 4:01:59 AM To: fvasluianu-cvl/SSSR Cc: Subscribed Subject: [fvasluianu-cvl/SSSR] Table 5 MaskShadowGAN[16] is 7.32 in ISTD and the result in USR (#1https://github.com/fvasluianu-cvl/SSSR/issues/1) Your work is great. ButI I have some questions about some content of the paper. [image]https://user-images.githubusercontent.com/23144348/97653868-bf33de80-1a9c-11eb-8c89-137a9b7dd2fb.png MaskShadowGAN[16] is 7.61 in ISTD (unpaired). Maybe 7.32 in Table5 is a small error. [image]https://user-images.githubusercontent.com/23144348/97653905-cf4bbe00-1a9c-11eb-9af6-530dc29048f5.png And I am confused about the result in USR test images (blue box). Because I don't find the same result in MaskShadowGAN[16]. MaskShadowGAN[16]: As the unpaired USR dataset has no ground truths, we conducted a user study to evaluate the shadow removal results. How do you get the results of the evaluation metrics, like RMSE, PSNR? Looking forward to your reply. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<#1https://github.com/fvasluianu-cvl/SSSR/issues/1>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AQCVKPMA5UJM5H7CMA4RY3DSNIUCPANCNFSM4TEPMEXQ.

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/fvasluianu-cvl/SSSR/issues/1#issuecomment-720445519, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AQCVKPISG5WMYEDK5HJQA23SN2RUFANCNFSM4TEPMEXQ.

nachifur commented 4 years ago

Yes, you are right in USR test image. But there are paired data in ISTD test images. According to your explanation, is it also a reconstructed RMSE on ISTD?