Description

Running Mask-RCNN on large scale video datasets could be time consuming. We are happy to release object masks predicted by Mask-RCNN on each frame for the VLOG and the EPIC Kitchen datasets. So far we are releasing object masks with a resolution of 100x100 and thresholded with a minimum confidence of 0.5. We hope to release higher resolution in a near future.

• VLOG masks (8.5G - 46G uncompressed)
• EPIC-AR masks (1G - 9G uncompressed)

Bibtex

In case of usage of our object masks predictions please cite our ECCV'2018 paper :

        @InProceedings{Baradel_2018_ECCV,
                author = {Baradel, Fabien and
                          Neverova, Natalia and
                          Wolf, Christian and
                          Mille, Julien and
                          Mori, Greg},
                title = {Object Level Visual Reasoning in Videos},
                booktitle = {ECCV},
                month = {June},
                year = {2018}
                }
        
    

Acknowledgements

This work was supported by the ANR/NSREC DeepVision project

Team

Fabien Baradel
INSA Lyon
Natalia Neverova
Facebook Research
Christian Wolf
INRIA - INSA Lyon
Julien Mille
INSA Centre Val de Loire
Greg Mori
Simon Fraser University