search

waps101 / depth-from-polarisation

Linear depth estimation from an uncalibrated, monocular polarisation image
Apache License 2.0
65 stars 29 forks source link

Error in height calculation #2

Open Slusny opened 3 years ago

Slusny commented 3 years ago

Dear Prof. Smith,

thank you for sharing your code and for the advances in polarimetric computer vision. I am currently working on my bachelor thesis, comparing your work against a multiview polarization method. Unfortunately, I have an issue with your height estimation. The result of your sample data and of my own data is always a linar increasing height with the x value. The actual depth of the object is not recovered. Until now, I haven't been able to fix this issue. I would really appreciate it, if you or someone from your team could help me, on why that might happen.

depthmap of the sample data: height_smith_croped

Wangyouai commented 2 years ago

[ height ] = HfPol( theta_est_combined,min(1,Iun_est),phi_est_combined,s,mask,azi,Bdist );same result as yours， [ height ] = HfPol( theta_est_combined,min(1,Iun_est),phi_est_combined,s,mask,false,spec )has better results. But it is still far from the original HfPol

Slusny commented 2 years ago

Hi Wangyouai,

Unfortunately, I wasn’t able to recreate the results. I ended up working with this paper from Dr. Zhu and Prof. Smith : https://github.com/AmosZhu/CVPR2019 , which was easier to get working correctly.

Best of Luck, Lennart

From: Wangyouai @.> Sent: Tuesday, 8 February 2022 07:31 To: waps101/depth-from-polarisation @.> Cc: lennytubby @.>; Author @.> Subject: Re: [waps101/depth-from-polarisation] Error in height calculation (#2)

Dear Prof. Smith,

thank you for sharing your code and for the advances in polarimetric computer vision. I am currently working on my bachelor thesis, comparing your work against a multiview polarization method. Unfortunately, I have an issue with your height estimation. The result of your sample data and of my own data is always a linar increasing height with the x value. The actual depth of the object is not recovered. Until now, I haven't been able to fix this issue. I would really appreciate it, if you or someone from your team could help me, on why that might happen.

depthmap of the sample data: https://user-images.githubusercontent.com/34861751/129046942-3e5bbd74-1d8d-4ffe-84cc-063a76a9579e.png

Hello, have you restored the experimental effect in the author's article? The depth information of my Demo results is not as good as the original paper.HlPol_Surface https://github.com/Wangyouai/Image/blob/main/2022-02-02%2022-08-05%E5%B1%8F%E5%B9%95%E6%88%AA%E5%9B%BE.png

— Reply to this email directly, view it on GitHub https://github.com/waps101/depth-from-polarisation/issues/2#issuecomment-1032257611 , or unsubscribe https://github.com/notifications/unsubscribe-auth/AIJ7FN54N3EZXIMCXZAZM2LU2C2C7ANCNFSM5B6VSKXA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . You are receiving this because you authored the thread. https://github.com/notifications/beacon/AIJ7FN2OHD57DNYXLAK3GW3U2C2C7A5CNFSM5B6VSKXKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOHWDQASY.gif Message ID: @.***>

Wangyouai commented 2 years ago

Hi Lennart ,

Thanks for the reply, I have tested the work of CVPR 2019 with Bayer polarized camera + realsense d435i, and the result is not ideal. May I ask what equipment you use to test it, if you can provide a set of data for me to test, I will be very grateful!

Regards, WangYA On 2/8/2022 @.***> wrote：

Hi Wangyouai,

Unfortunately, I wasn’t able to recreate the results. I ended up working with this paper from Dr. Zhu and Prof. Smith : https://github.com/AmosZhu/CVPR2019 , which was easier to get working correctly.

Best of Luck, Lennart

From: Wangyouai @.> Sent: Tuesday, 8 February 2022 07:31 To: waps101/depth-from-polarisation @.> Cc: lennytubby @.>; Author @.> Subject: Re: [waps101/depth-from-polarisation] Error in height calculation (#2)

Dear Prof. Smith,

thank you for sharing your code and for the advances in polarimetric computer vision. I am currently working on my bachelor thesis, comparing your work against a multiview polarization method. Unfortunately, I have an issue with your height estimation. The result of your sample data and of my own data is always a linar increasing height with the x value. The actual depth of the object is not recovered. Until now, I haven't been able to fix this issue. I would really appreciate it, if you or someone from your team could help me, on why that might happen.

depthmap of the sample data: https://user-images.githubusercontent.com/34861751/129046942-3e5bbd74-1d8d-4ffe-84cc-063a76a9579e.png

Hello, have you restored the experimental effect in the author's article? The depth information of my Demo results is not as good as the original paper.HlPol_Surface https://github.com/Wangyouai/Image/blob/main/2022-02-02%2022-08-05%E5%B1%8F%E5%B9%95%E6%88%AA%E5%9B%BE.png

— Reply to this email directly, view it on GitHub https://github.com/waps101/depth-from-polarisation/issues/2#issuecomment-1032257611 , or unsubscribe https://github.com/notifications/unsubscribe-auth/AIJ7FN54N3EZXIMCXZAZM2LU2C2C7ANCNFSM5B6VSKXA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . You are receiving this because you authored the thread. https://github.com/notifications/beacon/AIJ7FN2OHD57DNYXLAK3GW3U2C2C7A5CNFSM5B6VSKXKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOHWDQASY.gif Message ID: @.***>

— Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you commented.Message ID: @.***>

Chumumu commented 1 year ago

@Wangyouai @lennytubby I am reproducing this paper by Dr. Zhu and Professor Smith: https://github.com/AmosZhu/CVPR2019, in which I encountered some problems, I don't know how you solved it, in the normal correction part, after obtaining a rough depth map through stereo vision, how to get surface guidance normals from the depth map? This part is skipped in the demo code of the paper, and I think it is to apply the relationship between normals and gradients in the projected model to solve the normal, but I am not very sure and I hope you will explore this with me. My mailbox is a2468828528@163.com