1. Download the nyu_depth_v2_labeled_data.pkl directly, it contains all data in origin MATLAB version of Labeled Dataset. (I haven't upload it to this repo or to other network drives, it needs further clearing.)
2. Or you can follow the instruction below to generate it yourself (recommended now).

1. Download the Labeled Dataset from official website. And place that .mat file at the same dir.
2. Using the h5py module to load the .mat file.
3. The matrix in .mat file can be correctly and easily read, don't bother it.
4. The cell array in .mat file should specially processed.
5. The Map in .mat file should also be treated specially, I got its keys and values in MATLAB and opened them in MATLAB as a table format (can be firstly transposed in MATLAB), then copy the keys and values table to Excel to concatenate as one table with two columns. Finally, saved the Excel file as CSV format.
6. Above is how to treat different types of data field in .mat file. I just put all that field in one dict, and saved them as a pkl file. (~5.8GB)

The images and depth images can be easily visualized by matplotlib module.

• Maybe I will parse the something like the images out to a single directory, and just place string-like data in .pkl file.

@inproceedings{Silberman:ECCV12,
  author    = {Nathan Silberman, Derek Hoiem, Pushmeet Kohli and Rob Fergus},
  title     = {Indoor Segmentation and Support Inference from RGBD Images},
  booktitle = {ECCV},
  year      = {2012}
}