Image segmentation dataset generated from dash cam videos, consisting of 701 images (and their associated annotations. The dataset comprises 12 labels as opposed to the original CamVid dataset with 32.
Original dataset: SegNet-Tutorial-master.zip (179MB), labels
Conversion from original data:
Download the SegNet-Tutorial-master archive and extract it, resulting in the SegNet-Tutorial-master directory.
In the same directory, create the following sub-directories: grayscale, grayscale-split
-
Run this Python script (requirements.txt) in the top-level directory to create images/annotations in the grayscale and grayscale-split directories:
./venv/bin/python grayscale.py SegNet-Tutorial-master grayscale-split grayscale
The grayscale-split directory contains the train/val/test splits just like in the original data zip, but the dash cam image itself converted to JPG and the annotation in the same directory.
The grayscale directory contains all the images from the train/val/test directories, e.g., for converting it into other formats (e.g., bluechannel) or for generating different splits.
License
Unclear, like original CamVid dataset.