VOT2016 pixel-wise annotations

The "VOT2016 pixel-wise annotations" dataset contains pixel-wise per-frame annotations for sequences from VOT2016 dataset. The annotation is in a form of BW images for each sequence and each frame of the VOT2016 dataset, where values greater than zero mark pixels that belongs to the object. The data has been acquired in cooperation with Eyedea.

Release
The dataset can be downloaded here: VOTSeg dataset
Code that was used to generate rotated bboxes from the segmentation mask in VOT2016: bbox_fit



Publication
In case you use this dataset or the bbox fitting code, cite the following technical report:

@TechReport{Vojir-TR-2017-01,
author = {Vojir, Tomas and Matas, Jiri},
title = {Pixel-Wise Object Segmentations for the {VOT} 2016 Dataset},
institution = {Center for Machine Perception, K13133 FEE Czech Technical University},
address = {Prague, Czech Republic},
year = {2017},
month = {January},
type = {Research Report},
number = {CTU--CMP--2017--01},
issn = {1213-2365},
project = {TE01020415},
keywords = {Segmentation, VOT, Dataset},
}
Download links: paper, bibtex file