Depth estimation on custom images about monodepth

[mrharicot]

Hi,
Do you also have the corresponding input images?
It looks like there is a lot of distortion in these images.
Any data that looks too different from the training one will produce unreliable results, and given that we trained on at most 2 datasets it is fairly easy to find images that have different statistics.

[MeghaMaheshwari]

Let me send you the input images in sometime? Do you suggest that I should re-train the network with the new set of images. Is there any other suggestion in case the input images have different statistics compared to what you have trained on?

[MeghaMaheshwari]

Hey could you give me your email address? I would like to send u the images so that I can know what exactly the issue is.

[mrharicot]

Hi, you can find it on my personal page.
http://www0.cs.ucl.ac.uk/staff/C.Godard/

[MeghaMaheshwari]

Hi...I have sent you the images.

[MeghaMaheshwari]

Hi..I sent the images to [your ucl address] but the mail bounced back. Could you please help me verify if its the correct mail address?

[mrharicot]

Hi,
I did not receive it, but the email was correct (I redacted it to avoid spamming).
Can you try my gmail address? it's mr.haricot

[Gatsby23]

@MeghaMaheshwari Hey, can you help me how to develop the depth image ? I have use the focal and baseline to convert the disparities to the depth, but when I save it as a .png. The value has been converted to the RGB color form and can't get the depth value through imread(.png) in Matlab. So can you help me with that?