Should we generate them on our own using the provided run_coco_deeplab.py or they are ready for download somewhere?
Thanks
Posted by: divx @ Feb. 17, 2022, 11:37 p.m.Hello,
Considering that we are providing the links for the original datasets, and to be transparent with the semantic generation, you should use the run_coco_deeplab.py script. The validation folder already contains the segmentation maps.
Note that this track of the challenge aims at inpainting a region by using a semantic segmentation map. Nonetheless, the use of additional conditional information (eg., edges, object instances, etc.) will not be penalized. The only condition is to discuss and ablate such improvements.
Posted by: afromero @ Feb. 18, 2022, 9:26 a.m.