search

JiaRenChang / PSMNet

Pyramid Stereo Matching Network (CVPR2018)
MIT License
1.43k stars 423 forks source link

Disparity of right image #123

Open weihaosky opened 5 years ago

weihaosky commented 5 years ago

Hi, the output is the disparity map of the left image. Can I get the disparity of the right image? Many thanks!

dropletpeace commented 5 years ago

Hi, the output is the disparity map of the left image. Can I get the disparity of the right image? Many thanks!

Hi, I'm looking forward to get the right disparity map.Have you figured it out? Best wishes.

weihaosky commented 5 years ago

You can rotate the input image pair by 180 degree. Then the right image becomes left one.

WASCHMASCHINE commented 5 years ago

This will make learned biases apparent, i.e. hallucinating road surface like depths in the bottom sky.

dropletpeace commented 5 years ago

You can rotate the input image pair by 180 degree. Then the right image becomes left one.

Thank U very much for your replying, I tried to rotate the input image pair by 180° as you said, but the results seems not right like this: image

weihaosky commented 5 years ago

don't know why. It seems ok in my application.

dropletpeace commented 5 years ago

don't know why. It seems ok in my application.

When I exchange the input like this: left_fold = 'image_3(180)/' right_fold = 'image_2(180)/' It works very well. image Thanks for your kind help again.

weihaosky commented 5 years ago

This will make learned biases apparent, i.e. hallucinating road surface like depths in the bottom sky.

It shouldn't if the network do not estimate the disparity relying on over-fitting.