We worked with the Playing for data and Playing for benchmark datasets. For your own captures, you'll likely need to create your own annotations. You can use this code for that (keep in mind that the ground truth labels don't need to match fine-grained semantic classes, but rather coarsely different material types (like vegetation, road, sky, vehicles).
We worked with the Playing for data and Playing for benchmark datasets. For your own captures, you'll likely need to create your own annotations. You can use this code for that (keep in mind that the ground truth labels don't need to match fine-grained semantic classes, but rather coarsely different material types (like vegetation, road, sky, vehicles).