This repo has a bunch of different tools and data to help you get benchmarking on NYUv2 quickly. It includes a script to download and extract data, extracted segmentation labels and a pytorch dataset class.
Note that this repo only concerns itself with the fully labelled portion of the dataset. For the larger unlabelled version I would recommend checking out the tensorflow dataloader (https://www.tensorflow.org/datasets/catalog/nyu_depth_v2).
This repository contains everything needed to download and extract data for the NYUv2 dataset in python. Using 'extract_nyu_v2.py' you can download and extract samples consisting of:
- RGB images
- Depth images
- Surface normals
- 13 class segmentation labels
- 40 class segmentation labels
The surface normals are downloaded from here and everything else from here Additionally, you can avoid doing anything and simply download and use the pre-extracted segmentation labels found in:
- test_labels_13
- train_labels_13
- test_labels_40
- train_labels_40
This repo also provides a script for a pytorch dataset class which lets you get up and running with the NYUv2 dataset very quickly (Adapted from here). Simply copy the code in 'torch_nyuv2.py', initialize the dataset using the class, give it to a dataloader and off you go!
If you find this repo useful please consider giving it a star, so it can be more easily found by others. Also please feel free to propose changes and fixes or ask questions by creating an issue, include a @h-0-0 so that I'm notified.
Finally, thank you very much to ankurhanda, VainF and xapharius who's code I amalgamated to create this repo.