Support set selection

[daniel-cores]

If I understood correctly, you are selecting the support set for each image based on the category of the first annotation of this image. This is the code from inference_loader.py:

current_gt_class_id = int(gt_boxes[0][4])
selected_supports = self.support_pool[current_gt_class_id]

It seems that the support set only contains images from one category, which is the same as one of the annotations. Couldn't this image contain annotations from different categories? How can the method correctly detect these objects if the support set only contains images from one category?