search

sylqiu / tof_rgbd_processing

Off-the-shelf deep alignment and refinement for weakly calibrated ToF RGB-D modules
33 stars 7 forks source link

Questions about only train ToF-KPN #4

Closed Hongdou68 closed 4 years ago

Hongdou68 commented 4 years ago

Hi, sylqiu: I have read your paper, in your paper compapred with DeepToF is given ,however the input of the ToF-KPN is RGB image 、 wraped ToF amplitude and wraped ToF depth image, when you train a ToF-KPN without RGB image , how to get wraped ToF amplitude and wraped ToF depth image ?

Hongdou68 commented 4 years ago

beacuse there is no opetical flow network to estimate W_refn

sylqiu commented 4 years ago

Hi, thanks for your interest! The comparison is done only for the refinement module, and the Rgb input is removed. This setting is described in detail in the paper.

Cordialement,

Syl.

Envoyé de mon iPhone

On 16 Feb 2020, at 11:07, Hongdou68 notifications@github.com wrote:

 beacuse there is no opetical flow network to estimate W_refn

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

Hongdou68 commented 4 years ago

Hi, thanks for your reply, from your paper ,I know the RGB image is removed , But I did not understand how to get wrap ToF depth map and wrap ToF amplitude which are the input of the ToF-KPN? or you only use original ToF amplitude and de-aliasing ToF depth map for ToF-KPN

sylqiu commented 4 years ago

Do you mean warping for alignment? In that comparison there is no RGB involved, so no alignment is needed. The ground truth depth is originally aligned with the ToF depth.

Cordialement,

Syl.

Envoyé de mon iPhone

On 16 Feb 2020, at 12:41, Hongdou68 notifications@github.com wrote:

 Hi, thanks for your reply, from your paper ,I know the RGB image is removed , But I did not understand how to get wrap ToF depth map and wrap ToF amplitude which are the input of the ToF-KPN? or you only use original ToF amplitude and de-aliasing ToF depth map for ToF-KPN

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

Hongdou68 commented 4 years ago

Thanks，I understand it .And i have another question, from your paper , i have know how you genenrate the training data from the synthetic scene, but i have not found any program to do it, also do not known how to generate the synthetic scene. so do you have any code to do it?

sylqiu commented 4 years ago

Our data generation is based on Su’s deep end-to-end ToF imaging released code. You can find the code together with some scenes there.

Cordialement,

Syl.

Envoyé de mon iPhone

On 16 Feb 2020, at 14:43, Hongdou68 notifications@github.com wrote:

 Thanks，I understand it .And i have another question, from your paper , i have know how you genenrate the training data from the synthetic scene, but i have not found any program to do it, also do not known how to generate the synthetic scene. so do you have any code to do it?

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

Hongdou68 commented 4 years ago

okay, thank you very much for your help!!!