I'm looking for a dataset of natural images, with no specific requirement on the type of scene, designed for the task of detecting (automatically or visually) the presence and position of small objects. This could come for example from an eye-tracking experiment where participants are tasked to e.g. look for dogs in a picture taken with a camera, or an image segmentation dataset. The point is that the task is challenging/the category is rare/the segmentation area is small and hard to find. Each scene would ideally be associated to a label or set of labels indicating the object/s to be searched for, an explicit position or the segmentation masks are a plus but not required.