Note of caution
This is the beta version of the dataset, and might change in the near future. We will keep registered users up-to-date, but please make sure to check this page for the latest version.
What is in the archive?
This archive contains the beta version of the Sintel-segmentation training set. For each of the training sequences, it contains:
The segmentations
Here, we define a segment to be a set of pixels that (a) share the same material, and (b) belong to the same mesh. The labels of all segments are consistent across all frames of a given sequence, but not necessarily across different sequences.
A visualization of the segmentations is also included.
Masks of invalid pixels / segments
Due to aliasing effects, some segments are only a few pixels large. Therefore, we mask out all segments which are smaller than 25 pixels in all frames of a given sequence.
SDK
The SDK includes example scripts to read/write the segmentation data, and documentation on the data format.
How was it generated?
The dataset was generated by assigning unique IDs to each mesh and each material in Blender, and writing them out via the object and material passes. The final labels are a combination of these two unique IDs.
To generate the mask of invalid segments, we mark all segments for which the maximum size over the whole sequence never exceeds 25 pixel.
Examples
These example show the final pass with superimposed segment boundaries (top) and the color-coded segmentation (bottom).
Contact & citation
If you have any questions or problems regarding this dataset, please do not hesitate to contact us.
If you use this work, please cite:
@inproceedings{Butler:ECCV:2012,
title = {A naturalistic open source movie for optical flow evaluation},
author = {Butler, D. J. and Wulff, J. and Stanley, G. B. and Black, M. J.},
booktitle = {European Conf. on Computer Vision (ECCV)},
editor = {{A. Fitzgibbon et al. (Eds.)}},
publisher = {Springer-Verlag},
series = {Part IV, LNCS 7577},
month = oct,
pages = {611--625},
year = {2012}
}