Semantic Segmentation masks of Gibson Challenge Envs

Hello Team,

While trying to get Semantic Segmentation masks of the environments given in Gibson Challenge Datasets, using the sample code in examples/demos -, I found that the image obtained by selecting the ‘seg’ mode is all zeros. Does that mean that the mask does not exist or do I need to do something else?


Hi @Ruddra_dev, in static track the semantic segmentation masks are all zero, because it indicate everything belongs to the background class. In interactive/dynamic track, you can assign class_id to objects, which will generate non-zero values in semantic segmentation masks. Note that you don’t have access to semantic segmentation at test time, the state provided by our challenge only include RGBD images.


1 Like