search

mrharicot / monodepth

Unsupervised single image depth prediction with CNNs
Other
2.21k stars 628 forks source link

Test results not good after training on custom data #243

Open poornimajd opened 4 years ago

poornimajd commented 4 years ago

Hi, I have trained my model on Indian Driving dataset . But I get this as the output 0000000_disp 0000000 My data is similar to that of Kitti except that it is more challenging because it is on Indian roads , a unconstrained environment. I have around 8k left camera images and 8k right camera images which are in a sequence. Can you please help me out with this?

I have also attached a few examples from my training data 0000008 0000133

sieme97 commented 4 years ago

Hi, are your images rectified?

poornimajd commented 4 years ago

No they are not rectified. I just gave the left and right camera images.In my opinion rectifying images was not mentioned in the repo.

sieme97 commented 4 years ago

Images should be rectified.

poornimajd commented 4 years ago

After rectifying I will get one image as a result,then should I give this as the only input for both left and right side images?

sieme97 commented 4 years ago

You will get two images after rectification.

poornimajd commented 4 years ago

Okay I will try out this once.Thanks !