search

marcelampc / d3net_depth_estimation

Dense Deep Depth Estimation Network (D3-Net) in PyTorch.
118 stars 19 forks source link

ask for help #3

Closed wrongchen7 closed 6 years ago

wrongchen7 commented 6 years ago

How to use the code generate_blurred_dataset.m?

marcelampc commented 6 years ago

Hi! Here are the steps to make it work:

Dataset First you need to have some dataset with corresponding pairs of rgb and depth maps. You can download NYUv2 from here https://cs.nyu.edu/~silberman/datasets/nyu_depth_v2.html to perform same experiments as in papers. You can just download the Toolbox and the small split of the dataset (Labeled dataset (~2.8 GB)).

Preprocess data Depth maps must be aligned, cropped and invalid depth values (missing) must be filled-in (it's all in the NYUv2 Toolbox).

Blur Now, you just have to change lines 30-34 to define the paths to the rgb and to the depth maps and the destination paths for the resulting data (blurred rgb and same depth maps).

Code expects depth maps in millimeters and saves as uint16 also in milimeters.

Hope this helps!