search

xy-guo / Learning-Monocular-Depth-by-Stereo

Learning Monocular Depth by Distilling Cross-domain Stereo Networks, ECCV18
https://arxiv.org/abs/1808.06586
MIT License
94 stars 8 forks source link

Performance of `StereoNoFt` on Cityscapes #9

Closed kwea123 closed 5 years ago

kwea123 commented 5 years ago

In the paper you state "Stereo networks generalize much better and have smaller synthetic-to-real domain transfer problems."

When I tried release-StereoNoFt.ckpt on a Cityscapes stereo image pair, it produces the following result: (the image is resized to 1024x512 using cv2.INTER_AREA) image But on KITTI it's pretty good (image size 1280x384) image

The monocular model release-StereoUnsupFt-Mono-pt.ckpt has the same phenomenon: bad on Cityscapes but good on KITTI. I also tried removing the car hood but without much improvement. Could you please give us a guide on how to reproduce the result of Figure 7?

xy-guo commented 5 years ago

The precision is low for far away objects. I think you can try images which include more near objects like Fig.7?