lochenchou / dorn Goto Github PK

Hey there,

Thanks for your work. Does the code works as expected?

Hi, I trained your code from strach, but the validation result seems not good.

Your code use both image_02 and image_03, but the original paper seems only use one side of the camera, is this right?

Please tell me does the vliad result make sense?
Or I miss using some part of your code?
Or like NYU dataset, I need to do some processing for the downloaded data?

The KITTI data downloaded from KITTI/depth/depth_prediction

Thanks very much!

Hello, your implemention for Dorn on pytorch is very useful. While I can't train it, could you please provide the weight of your resnet?

The location of your code is in: DORN/model/resnet.py line:123
if pretrained: saved_state_dict = torch.load('/datasets/KITTI/depth_prediction/pretrained/resnet101-imagenet.pth', map_location="cpu")

Without this weight, the performance may not good, thanks very much.