Pseudo masks for cross-dataset evaluation

Dear authors,

Thanks for your great work. I have a question about the cross-dataset evaluation setting.
Can you clarify: about the pseudo gt masks you generated for cross-dataset evaluation, are they still on the training dataset, or on the target dataset?
For example, for the COCO->ADE20K experiment, are the pseudo masks on COCO (and all the 80 classes are already used for training?), or on the ADE20K (and when training the Mask RCNN you are essentially training on two datasets) ?
Thanks!!

Best,
Haiwen