About using the model.

[Cc-Hy]

Hello,
I wanna use this depth prediction model to construct my model on KITTI 3d detection task, specifically, i'd like it to generate a 4 times downsampled depth map with respect to the raw KITTI image(e.g. for an image of size 1242 * 375, i want to get a depth map of size 311 * 94), so i could simply use your pre-trained model by using
... --height 94 --width 311
Is that right?
Or is it better that i train a new model using
... --train --height 94 --width 311

[dg-enlens]

Hi Hy,

Unfortunately, the output image size is tied with the model. If you try to load our pre-trained model for a resolution different than --height 128 --width 416, you will probably get an error. You'd better redo a training at the desired resolution.

Also, please note that the --width and --height values must be multiple of 32. Your desired resolution of [311 x 94] won't work. You should probably select something like [320 x 96] instead.

Thanks